Home

Simulated Human being a dissertation submitted in partial fulfilment

1. END Hip 5 2 Joint Matrices Each joint has a 4x4 matrix which is the transformation that is to be applied to the mesh and all the children nodes This matrix is calculated as the product of all translations and rotations that have been applied on each node Instead of having these products done by the program there is an easier and more efficient way of doing them and that is to use OpenGL code OpenGL uses its own transformation matrices to translate rotate or scale the primitives The difference is that these products are executed on the hardware of the graphics card rather than the CPU which makes the program much more efficient The technique is to have OpenGL do all the transformations and then pop the 4x4 matrix from the stack and use it on the joint A sample code that demonstrates this technique is the following Void TreeNode transform glPushMatrix start a new transformation 56 Translat Translat an OO OO for int i 0 lRotatef lRotatef lRotatef f centre x centre y centre z rotation x 1 0f 0 0f 0 0f rotation y 0 0f 1 0f 0 0f rotation z 0 0f 0 0f 1 0f f centre x centre y centre z read the matrix from the stack 1GetFloatv GL_MODELVIEW_MATRIX matrix i lt childrenNum i for all children children transform glPopMatrix stop the transformation 6 DESCRIPTION OF THE CLASSES The program uses a large num
2. In the case of forward kinematics the animator has more and more transformations to control which while lending more freedom to achieve a more expressive animation may prove to be too complicated and intricate to achieve in practice On the other hand inverse kinematics algorithms have a high computational cost making it almost impossible to calculate joint angles for all degrees of freedom of a virtual human in real time In most cases a balance between these two approaches is desired 3 2 Keyframing Borrowing its name from a traditional hand animation technique keyframing requires that the animator specify critical or key positions for the objects The computer then fills in the missing frames by smoothly interpolating between those positions The specification of keyframes can be partially automated with techniques that assist in the placement of some body joints If the hand of a character must be in a particular location for instance the computer could calculate the appropriate elbow and shoulder angles using inverse kinematics Although such techniques simplify the process keyframing requires that the animator has a detailed understanding of how moving objects should behave over time as well as the talent to express that information through keyframed configurations The continued popularity of keyframing comes from the degree of control that it allows over the fine details of the motion 3 3 Motion Capture Motion
3. rotation has to be coordinated with the walking to achieve better visual results As seen in figure 4 7 when the body is rotating the human is standing on one leg and the whole body is rotating around it After that the leg is rotated as well When the human is walking and the body has to be rotated the program waits until one of the two legs is on the ground and the body is standing on it The body is then rotated around that leg so that the whole process is animated smoothly and realistically enough The state diagram for the walking process is shown in figure 4 8 Rotation around Left Leg Rotation around Right Leg Start rotating Figure 4 8 State diagram for the walking animation 5 PHYSICALLY BASED MODELLING 35 Traditional animators rely upon a mixture of visual memory and drawing skills to simulate various physical behaviours When a line test reveals that the animator is not correct it is redrawn and re tested until it satisfies the animator Unfortunately such techniques cannot be used in real time computer graphics If the aim is realism one must rely upon deterministic procedures that encode knowledge of the physical world These procedures are either based upon empirical laws or try to mimic a particular physical behaviour Simulations that are based on moving objects usually rely upon Newton s laws of motion Although they focus on the primary behaviours of theoretical particles having mas
4. Depending on the application a virtual human can have different roles as gt an actor all its actions are predefined by a script or other means The main application field is video film production an agent it acts as an autonomous entity and its actions depend on the environment Agents are usually found in simulations and virtual environments an avalar it serves as a visual representation of an actual person so its actions are guided in real time by that person The main application fields are virtual environments and video games The most important application fields that use or will use the current research on virtual humans ate Badler 1997 gt Engineering Design and Maintenance Analysis and simulation for virtual prototyping and simulation based design Design for access ease of repair safety tool clearance visibility and hazard avoidance Virtual Environments Virtual Conjerencing Efficient tele conferencing using virtual representations of participants to reduce transmission bandwidth requirements Living and working in a virtual place for visualization analysis training or just the experience Education Training Skill development team coordination and decision making Distance mentoring interactive assistance and personalized instruction Games and Entertainment Real time characters with actions and personality for fun and profit 18 Besides general industry driven
5. The independent rotation of different body segments is regarded as primitive motion A dialog will allow the user to change the attributes and create different postures of the human model Then the system will be tested on the accuracy and naturalness of the motion and if necessary the joint hierarchy and the human model will be adapted Definition of complex motion multiple joints The first task will be to define series of simultaneous transformations to different body parts and generate complex animations Then the physically based modelling component will be applied and the body muscles will be able to generate forces A number of different combinations will be tested to achieve certain tasks The first objective will be to apply forces in such a way that the body balance is maintained and then if possible more complex tasks involving motion will be tried Set up of a simple simulation The final aim will be to use the previous results in a simple application a simulation environment that will involve one or more virtual humans The nature of the simulation will depend on the achieved complexity in the previous steps 22 DESI GN 1 INTRODUCTION The design stage of this project has been a very important phase because of the problem s complexity and the number of design decisions that had to be taken The most basic part of the project has been the definition and modelling of a 3D environment The next stage was modellin
6. however as the requirements increase the problem becomes much more complex One example is the collision detection mechanism If an accurate collision with the objects is required e g for grasping an object the bounding cylinders might not be sufficient and more complex collision detection techniques should be used instead This would nevertheless reduce the efficiency of the program and it might need simpler models with lower polygon count to work in real time Another example is the collision response of the human limbs If a force is applied on one limb it will move it causing rotations on multiple joints These rotations are a complex inverse kinematics problem to calculate and in reality they also depend on the reaction forces applied on the muscles For example if a hand is pushed towards one direction the new joint angles depend on the muscle forces of the arm Some people may apply more force on the wrist others on the elbow etc Therefore an accurate human simulation may require a full biomechanical model with the ability to simulate muscle forces This approach will again increase the complexity of the program and reduce its efficiency in real time systems As a result one might state that there is no such thing as the perfect virtual human The success of the system depends strongly on the requirements and different approaches can serve different aims This project has been designed and implemented as a basis for real ti
7. slower computers may have fewer frames per second due to the inability of the hardware to render the scene in a time smaller than or equal to the timestep A frame rate greater than 30 frames sec is acceptable for animation but really smooth motion is achieved in rates over 60 frames sec The timer function updates the position and orientation of the scene elements including the camera and redraws the scene In this way the scene is refreshed at constant rates The timer s actions in a timestep are gt update the ball s position due to gravity and collisions gt update the virtual human s global position and the orientation of joints if walking is enabled gt update some joint rotations if a concurrent action is enabled gt redraw the scene 4 POSER MODELS AND VRML IMPORT 50 As stated in the design chapter the program should import the 3D geometry of virtual humans from files The implemented version supports VRML models created by Metacreations Poser Poser is an application especially built for generating and rendering human models therefore it is very suitable source for 3D geometry files of humans It can export the human meshes in various file formats such as 3ds 3D Studio dxf Autocad and wtl VRML 97 The latter has been chosen because VRML is fully documented and it has an ASCII file format which although more difficult to parse makes the testing and debugging of the file importer easier 4
8. target The speed of rotation change per joint depends on the success of the move If one segment is shaking the angle is increased in one step and decreased in the next or vice versa the joint speed is decreased to provide finer approach to the target On the other hand if an angle change is always heading towards the correct direction the speed is increased until it reaches the maximum value This procedural algorithm tries to mimic the real human behaviour In reality we do not know the exact target state of our arm when we try to reach for an object Starting from an initial state we continuously correct ourselves until we fulfil our target Additionally this algorithm is less complex than the common inverse kinematics solvers and can even work for moving targets It also takes the joint limits into account to avoid unnatural poses 43 One example of using the above mentioned algorithm is to have the virtual human reach for a ball In the case of fig 4 2 the joint chain would be Abdomen to allow the body to bend Left Shoulder Left Forearm The target function would return the distance between the Left Arm and the ball 44 IMPLEMENTATION 1 INTRODUCTION This chapter discusses some implementation issues of the project It describes the use of OpenGL and Visual C for the scene rendering and the interface analyses the way the VRML parser works and imports the virtual human geometry from files and describes
9. the constraints might state that the character should begin and end on the 15 ground and be in the air in the middle of the motion The optimisation software would then automatically determine that the character must bend its knees before jumping to get the maximum height for the minimum expenditure of energy Another approach finds the best control system by automatically searching among all the possibilities In the most general case this technique must determine how a character could move from every possible state to every other state Because this method solves a more general problem than that of finding a single optimum trajectory from a certain starting point to a particular goal it has been most successful in simple simulations and for problems that have many solutions thereby increasing the probability that the computer will find one Fully automatic techniques are preferable to those requiring manual design but researchers have not yet developed automatic methods that can generate behaviours for systems as complex as humans without significant prior knowledge of the movement Other researchers have tried to add emotions to simulated human motion Unuma 1995 Fourier expansions of experimental data of actual human behaviours serve as a basis from which this method can interpolate or extrapolate the human locomotion This means for instance that transition from a walk to a run is smoothly and realistically performed by the method For
10. 1 The VRML file importer The program is able to import any VRML file exported by Poser Nevertheless 1t does not have a generic VRML parser since this is outside of the scope of this project It exploits the way Poser generates the wrl files and can parse only these files The Virtual Reality Modelling Language is a file format for the description of 3D virtual worlds It was initially built to display 3D worlds over the Internet and was only supported by a few VRML plug ins Nowadays almost all 3D modelling and rendering applications support it and it has become one of the most common 3D file formats It is in ASCII form The most basic element of VRML is the Node A wrl file is actually a set of nodes and links between them Nodes have different types geometry lighting camera transformation etc and each node has a set of attributes The values of the attributes can be numbers arrays or other Nodes The most common and basic Node of VRML is Shape which describes the geometry of an object The geometry could be either a simple primitive such as sphere box cylinder etc or a mesh The Node that stands for a mesh is called IndexedFaceSet and is the one that has to be parsed from the ar file A sample Shape node in a file exported by Poser is the following CH EF Left_Thigh Shape appearance Appearance material Material geometry IndexedFaceSet coord Coordinate point 0 0024036 0 309619 0 0272996 0
11. Bird and Flock of Birds Aip www ascension tech com and the Polhemus Fastrak and Ultratrak btp www polbemus com home h m Sach systems are real time in that they can provide from 15 to 120 samples per second depending on the model and number of sensors of 6D data position and orientation with minimal transport delay 10 Figure 2 3 A performer in a typical configuration of magnetic motion capture sensors A typical magnetic motion capture system has one or more electronic control units into which the source and sensors ate cabled The electronic control units are in turn attached to a host computer through a network or serial port connection The motion capture or animation software communicates with these devices via a driver program The sensors are attached to the scene elements being tracked The source is set either above ot to the side of the active area There can be no metal in the active area since it can interfere with the motion capture The obvious solution for magnetic motion capture is to place one sensor at each joint However the physical limitations of the human body the arms must connect to the 11 shoulder etc allow an exact solution with significantly fewer sensors Because a magnetic system provides both position and orientation data it is possible to infer joint positions by knowing the limb lengths of the motion capture subject Optical Motion Capture Optical motion capture sys
12. KEYS AND MENU COMMANDS sssesssessseesseesseessressrtssresssressressressresseressressressreeseres 70 REFERENCES ageseent 72 ADDITIONAL BIBLIOGRAPHY 0 ccccssccsssescssscossosecesscccesscessosccsescessestessoscesoscesessdesssessoseeseescesesseces 75 111 INTRODUCTION 1 ABSTRACT The main subject of this dissertation is the design and implementation of a simulated human i e a three dimensional model that looks and acts like a real human Besides the review of the literature and the discussion about the most important issues concerning human modelling and simulation a project that demonstrates a virtual human with the ability to navigate and interact with its environment is presented The project has been designed and implemented so as to be reusable and extensible since its aim is not only to demonstrate some features of a simulated human but also to serve as a basis for future applications Such applications may include distributed virtual environments simulation systems etc 2 STRUCTURE OF THE DISSERTATION The dissertation is structured as follows In the next chapter all the background and related work is presented The focus is given on the modelling of the body and the animation of a virtual human Techniques such as keyframing motion capture and simulation are thoroughly described and compared to each other In the last part the main application areas of human modelling and simulation are briefly described The
13. Kingdom STEWART J and CREMER J 1992 Beyond keyframing An algorithmic approach to animation Graphics Interface 92 pp 273 281 Canadian Information Processing Society TANG W CAVAZZA M MOUNTAIN D and EARNSHAW R 1999 A constrained inverse kinematics technique for real time motion capture animation The Visual Computer 15 7 8 pp 413 425 WANG X and VERRIEST J 1998 A geometric algorithm to predict the arm reach posture for computer aided ergonomic evaluation The Journal of Visualization and Computer Animation 9 1 pp 33 47 WEBER L SMOLNIAR S W and BADLER N 1978 An architecture for the simulation of human movement ACM 78 Proc of the Annual Conf pp 737 745 WILLMERT K D 1975 Occupant model for human motion Computers amp Graphics 1 1 pp 123 128 WILLMERT K D 1978 Graphic display of buman motion ACM 78 Proc of the Annual Conf pp 715 719 78 WILLMERT K D and POTTER T E 1977 An Improved Human Display Model for Occupant Crash Simulation Programs Computers amp Graphics 2 pp 51 54 WOOTEN W and HODGINS J 1996 Animation of Human Diving Computer Graphics Forum 75 1 pp 3 14 ZHAO J and BADLER N 1994 Inverse Kinematics Positioning Using Nonlinear Programming for Highly Articulated Figures ACM Transactions on Graphics 134 pp 313 336 79
14. change should always follow the walking direction which can be determined directly from the global rotation of the body In the initial position the human body is facing the screen so its direction vector is 0 0 1 During any walking process the direction vector d is the rotation of the initial vector 0 0 1 by the global rotation angles If the virtual human is walking on a flat field the rotation around the y axis will determine its direction The state transition process presented so far is only capable of animating walking along a straight line However in most of the cases the virtual human may have to change its direction while walking This can not be done simply by interpolating the global rotation because the motion will not look natural In reality humans use their feet to rotate they can t just turn their bodies while walking The body rotation is a motion that has two states as shown in fig 4 7 These two states may not be enough because one cannot 34 usually rotate the body more than 90 degrees using just two steps If the rotation angle is bigger the rotation process is split into two smaller ones Figure 4 7 The two states for rotating the body If the human body is resting the rotation process can start immediately This is however not the case if the virtual human is walking If the rotation starts on arbitrary time one leg might be in the air and the animation will look unnatural Therefore the
15. chapter that follows is the project specification where the aims and objectives of the project are analysed Next is the design chapter where the design strategy and the theories and algorithms behind it are explained and justified It discusses about the definition of the 3D environment the body modelling the pose generation and human animation and the collision detection between an object and the human body The algorithms for major problems such as physically based modelling joint transformation inverse kinematics collision detection and response are described in detail The chapter that follows discusses the implementation of the program and the programming language and libraries that have been used The use of OpenGL and Microsoft Foundation Classes is explained the algorithms are presented from the programmer s point of view and all the implementation techniques are explained Problems such as the import of the human mesh from a VRML file the definition of joint structure and posture information files and the choice of data structures for the body skin and skeleton are discussed In the last part the techniques employed for testing the program are also described The implementation is followed by the results a chapter that presents the final program and its functionality The various options and commands of the program are explained and some screenshots are presented The program is not only treated as a standalone appl
16. contains corners edges and flat parts which will make the response of colliding objects look very strange For example if an object e g a ball collides with the human arm its response will not be predictable because it might hit a side of the bounding box an edge or a corner The reflected velocity after the collision will be a matter of luck This situation can be avoided if one uses primitives with a certain degree of symmetry such as ellpsoids Ellipsoids donot have corners and edges and they are symmetrical around one axis They have nevertheless other limitations Mathematically they are much more complex primitives compared to spheres and boxes making the collision detection much harder to compute in real time One good compromise between all the previous geometric primitives is the cy 4nder Fig 4 9 It is fairly symmetrical body limbs fit well in it and it is not very complex to calculate collision detection with it Therefore this solution has been adopted for the collision detection of the human body with its environment The bounding cylinders need to be calculated once and then they are only translated and rotated following the motion of the limbs they correspond to This calculation can be done initially when the human model is loaded The process is simple and follows the next three stages for each limb mesh of the body 1 Calculation of the bounding box One has to check all vertices of the mesh and find the hig
17. example the method gets briskness from the experimental data for a normal walk and a brisk walk Then the brisk run is generated by the method using another Fourier expansion of the measured data of running The superposition of these human behaviours is shown as an efficient technique for generating rich variations of human locomotions In addition step length speed and hip position during the locomotions are also modelled and then interactively controlled to get a desired animation Advantages and Disadvantages As a technique for synthesising human motion simulation has two potential advantages over keyframing and motion capture First simulations can easily be used to produce slightly different sequences while maintaining physical realism for example a person running at four meters per second instead of five Merely speeding up or slowing down the playback of another type of animation can spoil the naturalness of the motion Second real time simulations allow interactivity an important feature for virtual environments and video games in which artificial characters must respond to the actions of an actual person In contrast applications based on keyframing and motion capture select and modify motions from a precomputed library of movements 16 Although control systems are difficult to construct they are relatively easy to use An animator can execute a simulation to generate motion without possessing a d
18. improvements in the underlying computer and graphical display technologies themselves virtual humans will cause significant improvements in applications requiring personal and live participation In building models of virtual humans there are varying notions of virtual fidelity Understandably these are application dependent For example fidelity to human size capabilities and joint and strength limits are essential to some applications such as design evaluation On the other hand in games training and military simulations temporal fidelity real time behaviour is essential Badler 1999 Probably the most important application that utilises many aspects of human motion and simulation is a commercial system called Jack developed at the University of Pennsylvania Badler 1998 It contains kinematic and dynamic models of humans based on biomechanical data Furthermore it allows the interactive positioning of the body and has several built in behaviours including balance reaching and grasping walking and running The Jack model contains almost all the essential human skeletal joints and it can be scaled to different body sizes based on population data 19 PROJ ECT SPECI Fl CATION 1 AIMOF THE PROJECT There may be a considerable amount and variety of research going on in the field of human modelling and simulation but not all the aspects of the problem have yet been explored Applications like Jack mainly focus on the accur
19. is penetrating a cylinder and with these two primitives points and edges one can test boxes polygons or any kind of mesh 6 3 Collision Response If a collision with the cylinder has been detected the next step is to determine the collision response i e the new position and velocity of the object that has collided with it The general rule is that the velocity of the object after impact is equal to the reflection of the velocity before impact if the collision is perfectly elastic and there is no friction Fig 4 10 The reflection of the object s velocity can be found with the use of the normal vector of the cylinder s surface point with which the object has collided If the unit normal at the point of collision is N and the velocity on impact is v the velocity v r after impact will be v v 2 v NI N There are two cases 1 The object has collided with one of the two disks of the cylinder The normal vector is the vector of the plane on which the disk lies In the previous example the normal is either 100 or 1 0 0 depending on the disk 2 The object has collided with the cylinder s body The normal is the vector that results from the difference between the point of collision and its projection on the cylinder axis 41 In the previous example if px py pz is the point of collision the normal is the unit vector of 0 py cy pz cz Figure 4 10 The reflected velocity after the collision 7 D
20. joint J should also be applied to all other segments that follow in the hierarchy tree For example in fig 4 2 a rotation in the Right Thigh should also affect the Right Shin and the Right Foot Therefore the above mentioned process should be repeated for the vertices and normals of several other meshes This approach is not very efficient because segments that are in the lower positions of the hierarchy tree have to be rotated several times to determine the final vertex and normal values for a posture A more efficient solution is to use 4x4 matrices 30 A 4x4 matrix can both rotate and translate a vector when multiplied with 1t The general form of a transformation matrix is Pa Fa Fa x Nr Ty h Cy h x R and where Fa Fa a rotation matrix R and t t t a Mi Ta fa f fe pac ty 0001 P Za La translation vector T The question that arises is how to multiply a 4x4 matrix with 3 dimensional vectors The idea is to add a fourth dimension to the vectors which will also determine their type So instead of using x y z vertices will be in the form of x yz 1 and normals x y z 0 If one multiplies the transformation matrix with any four dimensional vector one will find out that if the fourth value is one the vector will be both translated and rotated in the case of zero it will only be rotated Therefore it is very practical to use 4x4 matrices and 4D vectors because the program does not need to determine i
21. of reaching for a particular point in space If the computerised character is much shorter than the captured motion the character must either reach up 13 higher to reach the same point in space changing the joint angles or reach to a lower point changing the position reached As differences become more complicated e g the computer character is the same height as the human model but has shorter arms so do the compromises in quality Geometric dissimilarity and motion stitching are two of the most difficult problems facing motion capture animators Solutions to these problems including inverse kinematics and constrained forward kinematics have had some success However these techniques require substantial user intervention and do not solve all problems 3 4 Simulation Unlike keyframing and motion capture simulation uses the laws of physics to generate motion of figures and other objects Virtual humans are usually represented as a collection of rigid body parts Although the models can be physically plausible they are nonetheless only an approximation of the human body A collection of rigid body parts ignores the movement of muscle mass relative to bone and although the shoulder is often modelled as a single joint with three degrees of freedom the human clavicle and scapula allow more complex motions such as shrugging Recently researchers have begun to build more complex models and the resulting simulations will become in
22. problem is to use fixed primitives in joints These primitives usually spheres or ellipsoids serve as a way to fill the cracks when the segments are moving giving the illusion of continuous skin The visual results of this technique are not as elegant as one might wish but due to its simplicity it has been adopted by many real time systems especially when less computational power is available One other more efficient solution is deformation by manipulation of skin contours Kalra 1998 This technique does not treat body segments as static meshes but deforms the body s geometry when a segment is rotated The idea is to manipulate the cross sectional contours set of points that lie on the same plane of the body mesh Every joint has to be associated with a contour and it has to lie on the plane of this contour When a segment is rotated around a joint the deformation of the skin around that joint is determined by the two adjacent joints the parent and the child in the joint hierarchy If the contour of the joint J that caused the rotation has normal N the normal of the plane on which the contour lies and the two adjacent joints J and J have normals N and N respectively the intermediate contours in the segments JJ and J J are calculated using linear interpolation between the normals Once the normal of a contour is calculated all its vertices are transformed so as to lie on the new plane This technique is fast and efficient bu
23. stance phase the ankle knee and hip of the leg that is in contact with the ground must provide support and balance When that leg is in the air however the hip has a different function that of swinging the limb forward in preparation for the next touchdown One state machine selects among the various roles of the hip and chooses the right action for the current phase of the running motion Associated with each phase are control laws that compute the desired angles for each of the joints of the simulated human body The control laws are equations that represent how each body part should move to accomplish its intended function in each phase of the motion To move the joints into the desired positions the control system computes the appropriate torques with equations that act like springs pulling the joints toward the desired angles In essence the equations are virtual muscles that move the various body parts into the right positions In recent years researchers have worked on simulation based methods that generate motion without requiring the construction of a handcrafted control system Several investigators have treated the synthesis of movement as a trajectory optimisation problem Witkin 1988 Gleicher 1997 This formulation treats the equations of motion and significant features of the desired action as constraints and finds the motion that expends the least amount of energy while satisfying those restrictions To simulate jumping
24. the general case an object possesses curved surfaces and the facets are an approximation to such a surface Polygons may contain an arbitrary vertex number ot be restricted to triangles It may be necessary to do the latter to gain optimal 23 performance from special purpose hardware or graphics accelerator cards In the simplest case a polygon mesh is a structure that consists of polygons represented by a list of linked x y z coordinates that are the polygon vertices Thus the information stored to describe an object is finally a list of points or vertices In all solid objects adjacent polygons share common vertices therefore it is a waste of memory to store the same vertices two three or more times in a mesh structure At first this waste may seem insignificant but one should not forget that with current technology some graphics scenes can have hundreds of thousands of polygons A more efficient approach is to have a list of all vertices stored in the mesh structure and have the polygons point to the vertices Some objects of the scene may contain curved surface parts and it is important for them to be rendered smoothly to add realism to the images This is necessary for applications containing virtual humans because the human body itself is full of curved parts Shading algorithms use the normal vector at the vertex of each polygon to create a smooth shading effect on a curved surface Therefore a normal vector should be assigned
25. transformations The fact that a change in the 5 rotation of one joint affects the position of several others makes it necessary to connect them in a tree structured topology Badler 1993 A change in the rotation of a node will cause geometric transformations on all its leaf nodes Recently an attempt has been made to standardise a joint hierarchy for Virtual Reality Modelling Language human models The Humanoid Animation Working Group H Anim proposed a specification for a standard VRML humanoid Roeh 1999 which includes a very detailed joint tree The primary goal of the working group is to encourage 3D artists to create skeleton models in a standard way so they can be used in different applications Sata ta Haka Cast dy L Gik pemn ai Visas Pais V Tiama k Figure 2 2 The joint hierarchy proposed by H Anim 2 3 Modelling the skin and clothes Careful consideration has to be given when modelling the human skin because unlike mechanical objects movement of the human body causes the surface to deform For example when the wheels of a car are turned this rotation does not affect any other parts of it but when the human arm is rotated it causes a deformation of the skin around 6 the shoulder and chest If the segments of the body are modelled as rigid objects even small rotations may cause unwanted cracks on the skin and make the model look unrealistic The simplest and easiest way to avoid this
26. 0023017 0 320267 0 0277575 51 0 0208637 0 328671 0 0431539 0 0359573 0 316325 0 0483461 coordIndex 2 31 4 1 78 1 8 10 1 VO OW Oo Aaa NEF normal Normal vector 0 879089 0 013376 0 47647 0 0736596 0 871212 0 48535 0 0792951 0 44513 0 891948 0 262508 0 0100572 0 964877 color Color color 0 470573 0 337255 0 239216 0 117647 0 0784314 0 0352941 0 105882 0 0666667 0 0313726 0 454902 0 188235 0 160784 colorPerVertex FALSI colorIndex Ej NRoo DEF Left_Thigh Shape The VRML files exported by Poser contain a set of such Shapes which define all the body parts Each shape has an IndexedFaceSet geometry The important attributes in the IndexedFaceSet definition that define the mesh geometry and have to be parsed by the program are 52 gt the point array in the Coordinate node it contains all the vertices of the mesh Every three numbers is an x y z set of a new vertex gt the coordlndex array it defines the polygons of the mesh Each polygon is a set of numbers terminated by 1 Each of these numbers is an index to a vertex in the Goin array The indices start from zero e g O 1 2 1 defines a triangle that connects the first second and third vertex of the point array gt the vector array in the Normal node it has the normal vectors of the mesh The normal vectors are not assigned to the polygons but to the ver
27. 46974 gt 0 064597 0 413107 0 049543 Chest Vertices 102 Colours 10 Faces 97 Bounding box 0 0799903 0 536267 0 0463849 0 463024 0 0578575 gt 0 0811718 58 Another important testing method which is common in graphical software was the visualisation of additional information in the form of geometric primitives For example in some stages the visualisation of the normals as 3D vectors and of the bounding cylinders in wireframe fig 4 9 was required Another important visual test was the comparison of the VRML model rendered by the program and by another VRML browser The physically based modelling mechanism was also tested and one appropriate way to do this was to test the reaction of the system in critical situations Some tests were included to have a ball collide with an edge or a corner of a primitive or to collide with two primitives simultaneously Another test was to have a perfectly elastic ball bounce around a room for a long time and check if there is gain or loss of total energy e g compare the maximum height of the ball over time The final test was the efficiency comparison where the frames per second of the program were measured in different hardware configurations This test showed that the program can work in acceptable frame rates in a common configuration but it is really efficient if the graphics card and CPU performance is higher than average 59 RESULTS 1 DESCRIPTI
28. A physically based modelling PBM component will be created to simulate the physics of the human body and the environment The human model will be imported from a 20 commercial 3D modelling package such as Metacreations Poser or Animation Master and a hierarchy tree will be designed to support the animation of the human body parts The next step will be to apply coordinated transformations in the human parts and generate complex and if possible goal oriented animation sequences such as walking or grasping an object Finally a simple simulation environment containing one or more instances of the virtual human will be set up Focus will be given on the functionality and extensibility of the implemented code to allow this project to be a basis for future work The stages of the project will be gt Implementation of a simple 3D environment An MFC application that will be the basis of the project s implemented system The environment will be able to load 3D polygons from a file and display them using OpenGL The implemented file format will support material diffuse colour specular colour and shininess for each polygon and transformations translation rotation scale and centre of rotation for sets of polygons Furthermore the user will be able to change the position and orientation of the viewpoint to visualise different aspects of the 3D world gt Physically based modelling A component able to simulate physical forces and
29. Computer Society MYSZKOWSKI K OKUNEV O KUNI T and Masumi Ibusuki 1997 Modelling of Human Jaw Motion in Sliding Contact The Journal of Visualization and Computer Animation 8 3 pp 147 163 PARKE F I and FRIEDELL M 1978 Interactive simulation of biomechanical systems the kinematics and stress of the human knee ACM 78 Proc of the Annual Conf pp 759 764 PHILLIPS C and BADLER N 1991 Interactive Behaviors for Bipedal Articulated Figures Computer Graphics Proceedings of SIGGRAPH 91 25 4 pp 359 362 Edited by Thomas W Sederberg TI POURAZAR G 1991 A method to capture and animate the dynamics of human motion COMPUGRAPHICS 91 I pp 181 197 1991 RAIBERT M and HODGINS J 1991 Animation of Dynamic Legged Locomotion Computer Graphics Proceedings of SIGGRAPH 91 25 4 pp 349 35 Edited by Thomas W Sederberg RIJPKEMA H and GIRARD M 1991 Computer animation of knowledge based human grasping Computer Graphics Proceedings of SIGGRAPH 91 pp 339 348 Edited by Thomas W Sederberg RIJPKEMA H and GIRARD M 1991 Computer animation of knowledge based human grasping Computer Graphics Proceedings of SIGGRAPH 91 25 4 pp 339 348 July 1991 Las Vegas Nevada Edited by Thomas W Sederberg STEWART J and CREMER J 1992 Animation of Human Locomotion Through Algorithmic Control Third Eurographics Workshop on Animation and Simulation September 1992 Cambridge United
30. DETECTION AND RESPONSE AEN 37 6E BOUAAME PLN iras 38 6 2 Collision Detecttion ETEA ES E TE 40 E GE e 41 7 DYNAMIC BEHAVIOUR eeunenn nnan a E a E ns 42 IMPLEMENTA TION tastscvsseccscecsasccsscccccssceciscsuisssusesscvasctedessdevestcecascevsscsuscssssesstedestevacscecctscectevevecdetesesvecss 45 1 INTRODUCTION EE 45 2 OPEN a dat 45 2 1 POLSON Render a A aaa 46 A ELA TE 46 3 VISUAL CAND MIRC AAA li 47 Jal MOUSE CON A ied Be Hes Baie ie 48 3 2 Keyboard Menus and Dielogs A 49 SS EE 50 4 POSER MODELS AND VRML IwMpoprt 50 4 1 The VRML filet iiip Overy iii a A beds 51 5 SKELETON DEFINITION AND MANIPULATION cccococccooonccoonnnconnncnoncconncnonnncnonnnnonnancnnnnnnonnncnnnncnnnnos 54 DL The Jont Hierarchy les A nee 55 A SOUL Matrices conn ess 56 6 DESCRIPTION OF THE CLASSES ui E E E A EEN 57 G TESTING tai 58 RESULTS cecascesdesctetectatvs eege eege ge 60 1 DESCRIPTION AND USE OF THE PROGRAM sss ssssssssssssesssessressressresssreseressressresssressressreesrresereseressee 60 DD e AAA E 60 1 2 Loading models and defining poettrerg 6l 13 Phy sic lly based modelling iii dE EAR EE EE ERER 62 1 4 Walking animation and other actions 62 IS Demonstration ER 63 Ze EXTENSBIE oia 64 CONCEUSTON Sa eege el 66 APPENDIX USER MANUAL csscccscccccccccccscsecscscccssscesssssesesoecss vecetsecsesscsessssescteceseescsseccvsecsusevecssvescssseed 68 1 CD CONTENTS ita ida 68 2 EXECUTING THE PROGRAM aia 69 3 SUMMARY OF
31. N 1997 Adapting Simulated Behaviors For New Characters SIGGRAPH 97 Proceedings Los Angeles CA HODGINS J and WOOTEN L 1998 Animating Human Athletes In Robotics Research The Eighth International Symposium Y Shirai and S Hirose eds Springer Verlag Berlin pp 356 367 HODGINS J 1996 Three Dimensional Human Running Proceedings of the IEEE Conference on Robotics and Automation 72 HODGINS J 1998 _Animating Human Motion Scientific American issue 03 98 http www sciam com 1998 0398issue 0398hodgins html accessed 9 March 2000 KALRA P MAGNETAT THALMANN N MOCCOZET L SANNIER G AUBEL A and THALMANN D 1998 Real time Animation of Realistic Virtual Humans IEEE Computer Graphics and Applications Vol 18 No 5 pp 42 55 LANDER J 1999 Devil in the Blue Faceted Dress Real time cloth animation Game Developer May 1999 pp 17 21 MAGNETAT THALMANN N CARION S COURCHESNE M VOLINO P WU Y 1996 Virtual Clothes Hair and Skin for Beautiful Top Models Computer Graphics International 96 Pohang Korea pp 132 141 O BRIEN J F BODENHEIMER B E BROSTOW G J and HODGINS J K 2000 Automatic Joint Parameter Estimation from Magnetic Motion Capture Data Proceedings of Graphics Interface 2000 Montreal Quebec Canada May 15 17 pp 53 60 POPOVIC Z and WITKIN A 1999 Physically Based Motion Transformation Proceedings of SIGGRAPH 99 Computer Graphics Proceedings Annual C
32. ON AND USE OF THE PROGRAM The program that results from this project is a windows application that demonstrates the use and functionality of the simulated human that has been designed and implemented as described in the previous chapters There is a 3D scene that displays a large room and a ball with the size of a football One can load various human models define and export postures and have them walking reaching for or kicking the ball Furthermore there is a full demonstration of the project s features with a male and a female model throwing the ball to each other 1 1 Camera control eg e Figure 6 1 The camera model The user has the ability to control the camera and rotate it in any direction using the mouse The camera is always pointing at a fixed point but it can rotate around it and zoom in or out If the user holds the left mouse button and moves the mouse the direction of the mouse defines the type of rotation Moving the mouse to the left or to 60 the right rotates the camera s position around the y axis Moving up or down similarly causes a rotation around the x axis The camera can also zoom in or out when holding the right mouse button and moving the mouse up or down respectively This camera model fig 6 1 is faster and easier to control compared to a full navigation model and it is also more suitable for examining the geometry of the human model and inspecting the various postures The program su
33. Simulated Human being a dissertation submitted in partial fulfilment of the requirements for the Degree of Master of Science in the University of Hull by Vosinakis Spyridon September 2000 Table of Contents INTRODUCTION mrenea s iere E E E SEE a E EEEE EAE Eoee ESSE 1 Ln ABSTRACT is 1 2 STRUCTURE OF THE DISSERTATION 1 BACKGROUND iicsesissscetss sesaseocessacsescncaseeststvedansusssbcscceesesessesenvedsdsecesodsdscgsescteds ide ed secsssnsaiecdsecatoueesseeessted 3 A INTRODUCTION ceci iia 3 Ze BODY MODELLING R EE T EE de 4 2 Modellina the EE 4 2 2 Modelling the skeleton iii ENEE ENEE REENEN vate 5 2 3 Modelling the skin and clothes srine a n a RRA S 6 235 HUMAN ANIMATION oenm rna aO E A EEGENEN RAAE EA 8 3 1 Body positioning Forward and inverse Enemoiiece 8 3 2 KEANU A a aaa 9 3 3 Moti n EE 10 IES TEE 14 LI Hybrid Approaches oia ti e a e e E E 17 d APPLICATION E RE 18 PROJECT SPECIFICATION cene rennene soaren aras 20 T CAIMOFTHE PROJECT oi A KEES 20 KE PROJECT STAGES E 20 A E BEE 23 Tu INTRODUCTION tia it 23 2 CREATING AN ENVIRONMENT ssc sscotscsaseseitacscnsesiaczssavestscagus oosuidenvend ses sqotastescucncni saben bdo 23 2 MODELLING THE BODY viii SEENEN 4 Bopy POSITIONING AND ANIMATION E NET 28 BL e 31 4 3 gt Ee 33 4A Walking animan On arsane O 34 5 PHYSICALLY BASED MODELLING sesecssesesesesesesesesesesesesesescesscesseaeseseseaeseseseseseeseseseseseaeaeaeeeaeeees 35 6 COLLISION
34. Visual C dll files The dynamic link libraries needed to execute the program Should be placed in c windows system folder or in the same folder as the executable Code Folder cpp files h files The program code with the necessary comments rest of the files The resource and workspace files to compile the program in Microsoft Visual C 68 Document Folder dissertation doc This dissertation in Word 97 format Table a 1 CD contents It is better not to execute the program directly from the CD but to copy the files to the hard drive All the files copied from the CD will have a read only attribute which has to be turned off if re compiling of the program is needed 2 EXECUTING THE PROGRAM The Program starts with the execution of the file Simulated Human exe which is located in the Program folder of the CD When the program is loaded the initial scene a large room with a ball is displayed The user can load a human model from the File Open menu The available models on the CD are one male and one female model man dat and woman dat Once the model is displayed the camera position can be changed by holding down the left mouse button and dragging the mouse to any direction The viewing mode can also be changed from the View menu There are three camera states Standard which focuses always on the centre of the screen Trace which follows the virtual human and De
35. YNAMIC BEHAVIOUR The term Dynamic Behaviour refers to actions executed by a virtual human that are not predefined but adapted to an ever changing environment This is therefore a very important issue for simulation systems because in most of the cases their objects creatures have to act in a dynamic environment Virtual humans with embedded dynamic behaviour demonstrate the most primitive form of intelligence i e the ability to decide on their actions according to the state of the environment Besides simulation this can also be an important feature of virtual environments Dynamic Behaviour could lead to a higher level of control and a more intelligent response of avatars as well as a more believable human like behaviour of virtual agents In this project the physically based modelling part inevitably generates a dynamic environment where objects obey gravity and the virtual human has to possess dynamic behaviour to interact with the environment The most primitive dynamic action is to follow a moving object e g a ball This action can be implemented by using the most basic principle of intelligent agents the look decide act scheme Obviously the look part is to read the coordinates position in space of the ball A more sophisticated version of it would be to be able to look at the ball only if it is within the virtual human s field of view but one might argue that the virtual human could a
36. acy of the simulation and have far too many details on human body structure because their main target is the field of industrial design and similar fields that need detailed results Therefore such systems run only in powerful workstations On the other hand applications such as Virtual Environments are not based on accuracy They require natural looking motion as well as acceptable execution speed and this is the aspect that this project will focus on The aim of this project is the design and implementation of a virtual human a human model that is able to execute actions and interact with its environment in near real time The results of this project can be extended in many ways First of all it can be used for photorealistic video production since all the motion data can be extracted where the animation will look much more realistic compared to the traditional approach of manually entering the motion sequence Then a real time simulated human can be also inserted in Virtual Environments such as 3D conference rooms virtual classrooms etc as synthetic actor and increase the believability of the virtual world Additionally complex simulations which involve human action e g fire emergency in a building can take advantage of the results of this project 2 PROJECT STAGES The project will be based on the C programming language and the OpenGL graphics library Microsoft Foundation Classes will be used for the windowing interface
37. all animation Animation Edit Demo Activates demonstration mode View Standard sets view mode to Standard View Trace sets view mode to Trace View Debug sets view mode to Debug View Eye sets view mode to Eye Help About information about the program Table a 2 Menu commands and Keys 71 REFERENCES BADLER N 1998 Jack description and features Online Available http www cis upenn edu hms jack html accessed 9 March 2000 BADLER N 1997 Virtual humans for animation ergonomics and simulation TERE Workshop on Non Rigid and Articulated Motion Puerto Rico BADLER N BINDIGANAVALE R BOURNE J ALLBECK J SHI J and PALMER M 1999 Real time virtual humans International Conference on Digital Media Futures Bradford UK BADLER N PHILLIPS C and WEBBER B 1993 Simulating Humans Computer Graphics Animation and Control Oxford University Press BROGAN D C METOYER R A and HODGINS J K 1998 Dynamically Simulated Characters in Virtual Environments IEEE Computer Graphics and Applications September October 1998 Volume 15 Number 5 p 58 69 DYER S MARTIN J and ZULAUE J 1995 Motion Capture White Paper http reality sgi com jam_sb mocap MoCapWP_v2 0 html accessed 9 March 2000 GLEICHER M 1997 Motion editing with spacetime constraints 1997 Symposium on Interactive 3D Graphics pp 139 148 ACM SIGGRAPH HODGINS J and POLLARD
38. apply them to the world s objects causing a change in one or more of their attributes position orientation etc The simulation will be executed in discrete time steps using the Euler method of numerical integration and besides physical forces such as friction collision and gravity it will also take into account those generated by the human muscles gt Definition of a joint hierarchy The human body will be divided into joints and segments and a geometric transformation tree will be designed Each body part will have certain degrees of freedom and specified limits in its movement The level of detail and complexity of the defined hierarchy will be carefully chosen since it will affect the accuracy the amount of computation needed to generate proper motion and the overall performance A good approach will be to allow more than one level of detail and to be able to switch between them Some joints will simply remain static will have fixed attributes if the detail level has been decreased gt Import of human model A primitive human model will be imported from a 3D modelling package and used as the basic model of the project The vertices and faces 21 of the 3D object will be converted to the implemented file format and the defined joint hierarchy will be applied to it Again the graphic detail of the human model will be such that it does not affect too much the overall performance Definition and testing of primitive motion
39. be projected using the values of the corresponding vertices of S These vertices that donot belong to S are marked as such and are not rotated when S is being rotated fig 4 5 The visual result is that all the polygons of S that are connected with S due to the common vertices are stretched or squeezed when the mesh is rotated having the effect of a continuous skin fig 4 6 32 Figure 4 6 Shoulder rotation without and with common vertices 4 3 Keyframing The design stage presented so far concerns the loading display and posturing of human models The next important state is the animation i e the coordinated motion of the body in real time The most usual animation technique is Aeyframing where the animator explicitly defines some key poses and the program interpolates between them Keyframing has been used in this project as the basis of animation sequences Let us suppose that the body has an initial posture P and the aim is to have a new posture P after time t Each posture can be described with the set of all joint rotations and the global translation of the body Let the joint J have initial t 0 rotation R rx ty rz and final t t rotation R rx ry rz Using linear interpolation between the two states the rotation of J at time t will be RI iers tex t t ty try t t ez tezi There is nevertheless a simpler way of calculating the interpolation In the beginning of
40. ber of classes which belong to one of these four categories Normal Abstract Template or MFC MFC classes are automatically created by Visual C when a new program is created or a new resource is being added The following table has a short description of all classes of the program The code is thoroughly described in the CD that accompanies this dissertation MFC classes Description CProjectl App The application class CProjectl Doc Contains document data and implements file saving and loading CProjectl View Implements the document display i e the scene rendering Dialog1 The dialog to change body part fig 5 1 Abstract classes Description String Static methods for string manipulation Streamable The streaming interface Template classes Description List A template to create double linked lists Node The node template of the list Normal classes Description Vector A 4 dimensional Vector Matrix33 A 3x3 Matrix Matrix44 A 4x4 Matrix Point A point of a mesh i e the vertex and the normal vector Colour The red green and blue values of a polygon Face A polygon of a mesh Contains an array of pointers to the points Mesh The mesh class that is used to manipulate and display a set of 57 polygons TreeNode A node of the joint hierarchy tree Person A virtual human as a set of meshes and TreeNodes fig 4 3 Cylinder The bounding cyli
41. bug which displays the bounding cylinders superimposed on the virtual human s body The Edit Change Active dialog from the program s menu allows the user to select an active body part of the human body and manipulate it Q and W keys rotate the active part around the x axis A and S around the y axis Z and X around the z axis If the user wants to store the current posture of the virtual human he she has to select the Edit Export Posture command from the menu The current joint angles will be stored in a file called posture dat Similarly loading a posture from the posture dat file and applying it to the virtual human is activated from the Edit Import Posture command The animation can be activated with the Edit Start Stop Animation command from the program s menu When the animation starts the ball obeys the laws of gravity and starts bouncing on the floor and keys start and stop the ball motion respectively If the ball hits the floor a wall or the virtual human it bounces off due to the collision detection and response mechanism of the program 69 The virtual human can be controlled from the keyboard using the cursor keys It walks forward with the up arrow and backwards with the down arrow Left and right arrows rotate the body 45 degrees to the given direction Holding down SHIFT and pressing the left or right arrow causes the virtual human to walk sideways to the left or right respectively The Space Bar at
42. capture involves measuring an object s position and orientation in physical space then recording that information in a computer usable form Once data is recorded animators can use it to control elements in a computer generated scene Animation which is based purely on motion capture uses the recorded positions and orientations of the body parts to generate the paths taken by synthetic objects within the computer generated scene Motion capture has proven to be an extremely useful technique for animating human and human like characters Motion capture data retain many of the subtle elements of a performer s style thereby making possible digital performances where the person s unique style is recognisable in the final product Because the basic motion is specified in real time by the subject being captured motion capture provides a powerful solution for applications where animations with the characteristic qualities of human motion must be generated quickly Real time capture techniques can be used to create immersive virtual environments for training and entertainment applications There are three different techniques that can be used to record the motion of the body magnetic systems optical systems and digital poseable mannequins Dyer 1995 Magnetic Motion Capture Magnetic motion capture systems use sensors to measure accurately the magnetic field created by a source Examples of magnetic motion capture systems include the Ascension
43. ces is known to allocate memory for the point array fig 4 1 and copy the vertices from the list to the array This approach is both efficient and not memory consuming 53 coordIndex array The program locates the coordIndex keyword It reads another string the character and then starts reading polygons until a P character is read Neither the number of polygons in the mesh is known a priori nor is the number of vertices in each face The parser uses the same technique as in the point array It stores all polygon pointers in an intermediate list and when the reading is over it allocates memory and copies them to the polygon array of the mesh For each polygon the parser reads indices until it reaches 1 For each index it finds the vertex s pointer from the vertex array and copies it to an intermediate list Again when 1 is reached memory is allocated and the list is copied to an array with pointers to the vertices vector array The program locates the vector keyword and reads the normal vectors Each vector is assigned to the appropriate vertex in the point array of the mesh color array The colours that will be used in each mesh can vary in number so an intermediate dynamic list is used too The program locates the color keyword reads the RGB value of the colours and stores them in the list Once the size is known it allocates memory for the colour array and copies the colours When the rea
44. creasingly lifelike as researchers continue to add such detail Hodgins 1998 Kalra 1998 Figure 2 4 Simulated Human Running GeorgiaTech College of Computing When the models are of inanimate objects such as clothing or water the computer can determine their movements by making them obey equations of motion derived from 14 physical laws In the case of a ball rolling down a hill the simulation could calculate motion by taking into account gravity and forces such as friction that result from the contact between the ball and the ground But people have internal sources of energy and are not merely passive or inanimate objects Virtual humans therefore require a source of muscle or motor commands a control system This software computes and applies torques at each joint of the simulated body to enable the character to perform the desired action A control system for jogging for instance must determine the torques necessary to swing the leg forward before touchdown to prevent the runner from tripping Hodgins 1998 Most control systems use state machines algorithms implemented in software that determine what each joint should be doing at every moment and then ensure that the joints perform those functions at appropriate times Hodgins 1996 Running for example is a cyclic activity that alternates between a stance phase when one leg is providing support and a flight phase when neither foot is on the ground During the
45. ct did not move during dt find the position p of the object exactly when the collision occurred Find the relation A between the impact position and the current one to approximate the exact time of impact and the object s velocity at that time In Po ES IP Po Ab lt 1 EE dt 4A dt 6 Assign the position p and velocity v to the object and repeat the process go to step 1 for the rest of the time dt dt 6 COLLISION DETECTION AND RESPONSE 37 In a simulated environment all objects have to behave as rigid bodies i e no object should be allowed to penetrate another The process of collision detection is to examine if the whole or part of the geometry of an object is within another Associated with this is another equally important process that of collision response which should determine the new position and velocity of the two or more objects that collided Calculating an accurate collision detection and response for a human body mesh is not an easy task One has to check each polygon of the mesh against all the other objects of the scene and determine if it is penetrating another polygon This is a very slow process especially for scenes containing virtual humans where the total number of polygons will be some thousands There are of course acceleration techniques that can be used to prevent testing collision against a other polygons like the use of bounding boxes but polygon polygon collision det
46. dation Class MFC Library 47 Application Framework as a C interface to the Windows API MFC which is fully supported by Microsoft Visual C is both efficient and relatively easy to use Some of the features of MFC which indicate why it is a very important tool for windows based applications are gt Full support for File Open Save and Save As menu items and the most recently used file list gt Print preview and printer support gt Full support for the common controls in the released version of Windows95 gt Classes for thread synchronisation The use of MFC in this project resulted in a rich application interface The use of keyboard and mouse input menus and dialogs gives the user the ability to have a better control of the program The mouse is used to control the camera the keyboard to move the virtual human and select different actions and the menus to use the various options of the program e g file import export and switch between different states e g camera types 3 1 Mouse control The mouse input has been handled with three different functions gt OnLButtonDown this function is called once when the user presses the left mouse button gt OnLButtonUp this function is called once when the user releases the left mouse button gt OnRButtonDown this function is called once when the user presses the right mouse button gt OnRButtonUp this function is called once when the user r
47. demonstration mode 2 EXTENSIBILITY Unless it is a full standalone application a project is of no value if future projects cannot take any advantage of it Therefore the aim of this project is not only to demonstrate a simulated human and the set of actions it can support but also to make the programming code as extensible and reuseable as possible This feature is of major importance because it will enable OpenGL based graphical applications easily to embed virtual humans in their scenes A great deal of focus was given on extensibility and reusability while this project was being designed and implemented One important step in that direction was the fact that the program directly supports VRML models exported from Poser Each model may need modifications on the joint structure and centres of rotation but it remains a fact that one can use any of the numerous Poser models in the program The file importer may not be a generic VRML parser but since it can read Mesh information the IndexedFaceSet node one could easily adjust any VRML file to make it readable by the program Almost all 3D file formats can be converted to VRML so with slight 64 modifications one could take advantage of almost any 3D geometrical figure available and use it on the program The only drawback will be that one will have to manually generate the joint hierarchy file One other important feature that enhances the extensibility of the project is the obj
48. ding of these arrays is over the program tries to find the next Shape node until the end of file is reached When a shape node is located a new mesh object is created and its values are read from the file The defined name of the Shape in the VRML file DEF name Shape is stored in the name attribute of the Mesh class 5 SKELETON DEFINITION AND MANIPULATION As stated in the design chapter the body skeleton is a tree structure of joints This structure is represented in memory with the use of pointers Each TreeNode object has an array of pointers to other TreeNode objects and each one of these points to others etc Each node should have exactly one parent node and a number of child nodes except for the head that does not have a parent node and the leaves that do not have child nodes 54 5 1 The Joint Hierarchy files The TreeNode hierarchy is read from another file the joznt hierarchy file This is the basic file that the program reads to load a virtual human It starts with the name of the VRML file that contains the human geometry and describes the tree of joints that will be used for posturing and animation The specification of the joint hierarchy file is as follows MESH path of VRML file NODE name of the node centre x yz min H max H xyz xyz an arbitrary number of children nodes END The name of each node should be the same as the name of the mesh Shape Node in the respective VRML fi
49. ect oriented design and implementation of it A proper object oriented methodology ensures a code more reusable and easier to debug Any programmer that may need to use some features of this project for a future application can easily include a subset of classes without worrying too much about the subtle implementation details One does only need to use the interface correctly and call the appropriate methods The most important class is the Person class which has the high level control over the virtual human Let us suppose that one application needs to have human bodies as avatars e g a virtual environment for collaboration over the Internet This application will have to include the Person class and all other classes that are directly associated with it The use of this class will be as simple as the following commands Command Description Person p create a person object file gt gt p load a human model from a file p initWalk start stop walking p stopWalk p initRotation float angle rotate the body in a given angle p prepareDraw apply all translations and rotations on the joints and create the final polygons to be drawn p draw render the mesh on the screen Table 6 1 Sample commands to embed a virtual human in an OpenGL application 65 CONCLUSIONS The main conclusion from this project is that it is actually possible to implement a class of simple but practical simulated humans
50. ection is still a very inefficient process for real time systems 6 1 Bounding Primitives There are a number of other collision detection techniques that are more efficient than polygon polygon collision detection but as expected less accurate Nevertheless the primary aim of real time systems is not accuracy It is true that the more accurate the calculations are the more believable the visual results will be but efficiency is still the most important issue One interesting technique is to compare bounding geometrical primitives of groups of polygons instead of comparing the polygons themselves The most used primitives are spheres and boxes mainly due to their simplicity which makes penetration tests much easier Having one giant bounding box or sphere around the entire human body mesh is far from being a good solution to the collision detection problem If an object were near the human body it would collide with an invisible box or sphere A much better approach is to have one bounding primitive per body limb The use of bounding spheres is still not a good solution because a sphere is equally large in all dimensions which is not the case in most of the human limbs Only the head and hands would fit quite well in a bounding sphere The bounding box is a much better approach since all the limbs fit very well inside boxes but there is one important drawback the collision response The box has the 38 problem that 1t
51. ed Motions with Relative Paths and Synchronization Graphics Interface 95 pp 18 25 Canadian Human Computer Communications Society Edited by Wayne A Davis and Przemyslaw Prusinkiewicz 76 LEE J and WOHN K 1994 Fuzzy Aggregation of Motion Factors for Human Motion Generation Virtual Reality Software and Technology Proceedings of VRST 94 August 23 26 1994 Singapore pp 55 70 World Scientific Publishing LEE P WEI J and BADLER N 1990 Strength guided Motion Computer Graphics 22 4 August 1990 LO J and METAXAS D 1999 Recursive Dynamics and Optimal Control Techniques for Human Motion Planning Computer Animation 99 May 1999 Geneva Switzerland IEEE Computer Society LOIZIDOU S 1992 Hidds Hybrid Inverse and Direct Dynamics System for Human Animation Third Eurographics Workshop on Animation and Simulation September 1992 Cambridge United Kingdom MAHMUD S and ZGUC B 1991 Semi Goal Directed Animation a new abstraction of motion specification in parametric Rey frame animation of human motion Eurographics Workshop on Animation and Simulation pp 75 87 MULTON F and ARNALDI B 1999 Human Motion Coordination Example of a Juggler Computer Animation 99 IEEE Computer Society MULTON F NOUGARET J HEGRON G MILLET L and ARNALDI B 1999 A Software System to Carry out Virtual Experiments on Human Motion Computer Animation 99 May 1999 Geneva Switzerland IEEE
52. eleases the right mouse button gt OnMouseMove this function is called when the user is moving the mouse 48 The mouse is used in the project to rotate the camera and zoom in or out The rotation is achieved while the user is holding the left mouse button and moves the mouse If the mouse is moved to the left or right x axis the world is rotated around the y axis A move up or down y axis rotates the world around the x axis Respectively zooming in or out requires the user to hold the right button and move the mouse To achieve this effect the following technique has been used gt There is a variable which can have three states ROT if the world is being rotated ZOOM if the camera zooms in or out and NOOP if none of the above happens gt Inthe OnLButtonDown function ROT state is activated gt Inthe OnRButtonDown function ZOOM state is activated gt Inthe OnLButtonUp and OnRButtonUp functions NOOP state is activated gt Inthe OnMouseMove function the system checks which is the current state If it is NOOP it saves the mouse position and exits If it is ZOOM it checks the distance between current mouse position and the previous one and zooms in or out in proportion to the y distance between the two mouse positions If it is ROT it rotates the world around the y axis proportionally to the x distance between the two mouse positions and around the x axis proportionally to the y distance 3 2 Keyboard Menu
53. equest or by streaming The joints on the monkey may be locked or tightened individually allowing adjustment of single joints This implies a painstaking though familiar workflow for the digital stop motion animator Each frame or keyframe of an animation sequence must be adjusted precisely then recorded using a motion capture program Disadvantages of Motion Capture Motion capture may have many advantages and commercial systems are improving rapidly but the technology has drawbacks Both optical and magnetic systems suffer from sensor noise and require careful calibration O Brien 2000 Additionally measurements such as limb lengths or the offsets between the sensors and the joints are often required This information is usually gathered by measuring the subject in a reference pose but hand measurement is tedious and prone to error Because of constraints on mismatched geometry quality constraints of motion capture data and creative requirements animation rarely is purely motion capture based In most of the cases animators have manually to alter the data to reach the desired results To maintain the integrity of motion captured data the scene elements being controlled by the data should be as geometrically similar as possible Depending on the degree that the geometries are different some compromises have to be made Essentially either angles or positions can be preserved though typically not both A simple example is that
54. etailed understanding of the behaviour or of the underlying equations Simulation enables control over the general action but not the subtle details For instance the animator can dictate the path for a bicycle but cannot easily specify that the cyclist should be riding with a cheerful light hearted style This limitation could be overcome in part by using simulation to generate the gross movements automatically and then relying on keyframing or on motion capture for the finer motions such as facial expressions One drawback of simulation is the expertise and time required to handcraft the appropriate control systems Another disadvantage is the computational cost Virtual environments with dynamically simulated actors need multiple processors with either virtual or physical shared memory to obtain the required performance Brogan 1998 Dynamic simulation also imposes some limitations on the behaviour of synthetic characters Simulated characters are less manoeuvrable than those modelled as point mass systems and those that move along paths specified by animators For example although a point mass model can change direction instantaneously a legged system can change direction only with a foot planted on the ground If the desired direction of travel changes abruptly the legged system may lose its balance and fall These limitations although physically realistic and therefore intuitive to the user make it more difficult to design robust algo
55. f a vector is vertex or normal They will all be multiplied by the same matrix Each time a segment has to be rotated its vertices and normals will be multiplied by a 4x4 matrix This single matrix will be the product of all the transformations that should have been applied to the segment For example if the segment S has to be rotated by the joint J its normals and vertices will be multiplied by the matrix M where 1 0 0 px 1 0 0 px 0 1 0 py 0 1 0 M PY RRR PY 0 0 1 pz 0 0 1 pz 000 1 000 1 For each joint J one can find the corresponding local transformation matrix M which should be applied to the descending segments Each segment s final coordinates depend on the transformations from all previous joints The final transformation is determined by a single 4x4 matrix which is the product of all corresponding matrices of the previous joints in the hierarchy tree For example in the hierarchy of fig 4 2 the final transformation matrix of the Right Foot mesh will be M M M M if MI M and M ate the local transformation matrices of the Hip the Right Thigh and the Right Shin respectively 4 2 Continuous skin 31 Let us suppose that the segments S and S are connected through the joint J If J is rotated S will be rotated but S will not This will cause a crack around the joint as already described in the Background chapter There are many ways to deal with this problem and the simplest one i
56. g of the body and skeleton of the virtual human followed by the positioning of the body and animation stages After that physically based modelling and collision detection had to be defined while the virtual human had to demonstrate some dynamic behaviour as well 2 CREATING AN ENVIRONMENT The graphical environment is the basis of this project It is the context in which a virtual human exists the means to visualise its appearance and actions This effort aims to display graphics that are realistic enough to be used in virtual environments or other real time simulations therefore the use of a three dimensional environment is essential There are a number of different approaches to generating three dimensional objects but currently the most efficient one for real time applications is the use of 3D Polygons This is mainly because they are supported by all hardware accelerated graphics cards making the rendering of a scene a very fast process with relatively good quality Additionally creating polygonal objects is straightforward and visually effective algorithms exist to produce shaded versions of objects represented in this way All other approaches have to make use of the CPU to visualise the virtual space thus increasing the rendering time dramatically and making it almost impossible to render complex scenes in real time In the polygonal approach each three dimensional object is represented by a mesh of polygonal facets In
57. h animation using a simplified spring mass model Lander 1999 3 HUMAN ANIMATION People are skilled at perceiving the subtle details of human motion A person can for example often recognise friends at a distance purely from their walk Because of this ability people have high standards for animations that feature humans For computer generated motion to be realistic and compelling the virtual actors must move with a natural looking style Specifying movement to a computer is surprisingly hard Even a simple bouncing ball can be difficult to animate convincingly in part because people quickly pick out action that is unnatural or implausible without necessarily knowing exactly what is wrong Animation of a human is especially time consuming because numerous details of the motion must be captured to convey personality and mood The techniques for computer animation fall into three basic categories Aeyframing motion capture and simulation Hodgins 1998 All three involve a trade off between the level of control that the animator has over the fine details of the motion and the amount of work that the computer does on its own Keyframing allows fine control but it is the animator who has to ensure the naturalness of the result Motion capture and simulation generate motion in a fairly automatic fashion but offer little opportunity for fine tuning 3 1 Body positioning Forward and inverse kinematics A body posture can be gene
58. hat connects the joints With this structure one can identify the joints and segments that are going to be affected if a joint is rotated For example a raise in the arm affects the position of the wrist and the hand As stated before all meshes have a unique name which is also the name of the segment they represent To avoid introducing new names for the joints each joint has the name of the segment that it directly affects For example the name of the wrist joint is hand 26 gt the joint position The joints have a maximum of three degrees of freedom they allow only rotations Therefore each joint has a fixed position that is also the centre of rotation for all its descendants in the tree structure v the joint limits A real human cannot rotate his her limbs arbitrarily in all directions every joint has certain limits These limits are introduced in the virtual human s joints as six numbers the minimum and the maximum rotation angle per degree of freedom The joint hierarchy tree that is being used in the default models of the project is shown in the Figure 4 2 Left Thigh Abdomen Right Thigh Left Shin Chest Right Shin Left Foot Left Collar Neck Right Collar Right Foot Left Shoulder Head Right Shoulder 9 Left Forearm Right Forearm Left Hand Right Hand Figure 4 2 A sample joint tree As stated before the skeletal information is a tree structure Each node of the tree which is actually a joint
59. hest and lowest x y and z values The result will be two vectors low and high which will define the corners of the bounding box The size and centre of the box can be found very easily size high low and centre high low 2 2 Cylinder type Each cylinder can be aligned on the x y or z axis The use of arbitrary cylinders has been avoided because it would complicate the calculations significantly This is not a drawback since they can be rotated and translated as any other object in the scene it just affects their initial definition The axis on which the cylinder will be aligned will be the one with the highest value in the size vector of the bounding box 3 Centre radius and height The cylinder s centre is the centre of the bounding box Its height or length or width depending on the axis is the size value of the axis on which it is aligned The cylinder s diameter is the mean between the size values of the two other axes For example if the cylinder is aligned on the x axis and the size 39 vector of the bounding box is sx sy sz then the length will be sx and the diameter me So the radius is going to be A Figure 4 9 A human mesh covered by bounding cylinders 6 2 Collision Detection After defining the initial position and size of each cylinder the next step is to check for collision with other objects in the scene In each timestep the cylinders have the translation and rotation of the limbs they c
60. ication but also as a piece of code that can be extended and reused in future applications The second part of the chapter explores this side of the program In the last chapter of this dissertation are the conclusions There is a comparison between the aims and results of this project and a discussion about the difficulties of this application area and the value of the project for graphics applications The dissertation closes with a description of possible future extensions After the conclusions there is an Appendix which is the user manual followed by the References and the additional bibliography BACKGROUND 1 INTRODUCTION Simulated or virtual humans are computer models that look and act like real humans They can vaty in detail and complexity and can be applied in fields such as engineering virtual environments education entertainment etc They ate also used as substitutes for real humans in ergonomic evaluations of computer based designs or for embedding real time representations of live participants into virtual environments Badler 1999 Human motion interaction and simulation especially for cases where an accurate model is required are extremely complex problems Therefore human simulation is a field of continuous research and many different approaches have taken place Accurate human motion requires complex physically based modelling where the body should not be treated as a passive object but as a set of joi
61. is stored in a class called TreeNode This class has the following properties gt the position of the joint v the three rotation angles of the joint v a pointer to the mesh that it directly affects v an array of pointers to the descendant nodes The assignment of a mesh to a TreeNode should be done in the beginning of the program by comparing the name property of a mesh with the joint name The head of the hierarchy tree which is usually the Hip node defines the global translation and rotation of the body and can of course access the whole set of joints of the virtual human In the general case where one application uses more than one virtual human there should be a Person class with a pointer to the head of the hierarchy tree The Person class should also have the entire set of meshes that define the body of the virtual human The class diagram for these classes is the following Fig 4 3 Figure 4 3 The class diagram of the Person Mesh and Treenode classes 4 BODY POSITIONING AND ANIMATION 4 1 Joint rotations When the human model is loaded all joint angles are zero The posture of the model is its initial pose and every other transformation is applied on this pose This means that every posture is actually a set of rotations on the initial pose Although nothing prevents a model from having any arbitrary pose as the initial one the same pose will look entirely different on two different models if their initia
62. ject was actually the process of scanning all the mesh objects of the scene and for each mesh displaying all its polygons using the previous piece of code The only additional code was the definition of the lights position direction and intensity and the viewing transformation that defined the distance and angle of the camera 2 2 Camera control OpenGL renders the scene from a fixed viewpoint and direction so the only way to define a camera transformation is to inverse transform the world For example if the camera is at the point P and has a viewing angle Ox Oy 02 in OpenGL one will have to translate the world by P and then rotate it by CDs Oy Dei This can be expressed in OpenGL code as follows 46 lPushMatrix lRotatef 0x 1 lRotatef 0y 0 0 0 0 rotation around y axis lRotatef 0z 0 0 0 0 1 0 rotation around z axis lTranslatef px py pz translation ender a function that renders the world 1PopMatrix 0 0 0 0 0 rotation around x axis Dijo ek Q KAA aQ One other way of defining the camera is to have its focus fixed on a certain point P of the world and rotate it around that point This is very practical in cases like examining an object where the fixed point should be the object s centre In such cases the world first has to be rotated around the point P and then translated to the desired distance of the camera D The appropriate OpenGL code would be lTra
63. l one is not the same Therefore it is necessary that all models start from the same pose if the predefined animation is going to be applied to them The most common initial pose among the programs protocols that animate virtual humans and the one used in this program is the following Fig 4 4 28 Figure 4 4 A sample initial position for virtual human Each pose is a set of rotations applied on joints For example a raise in the arm is one rotation applied on the shoulder joint There are two ways to define a rotation using a rotation axis and one angle or using three angles one per axis The second approach has been adopted mainly due to its simplicity in both defining a rotation and manipulating it Each of the three rotation angles is compared to the joint limits the maximum and minimum angle value per axis and the program determines if it is valid or not Let us suppose that the joint J has a rotation R rx ry rz and a position P px py pz L The joint rotates the segment S and all its descendants in the hierarchy tree The vertices of S are defined in absolute coordinates world coordinates and not in terms of the joint centre P This means that they cannot be directly rotated by R because after the rotation S will not be attached to J anymore The rotation should be around P which is the joint centre and thus the centre of rotation for S and all its descendant nodes The process of rotating a mesh is act
64. le The centre value is the centre of rotation of the joint and the min and max values are the joint constraints i e the minimum and maximum rotation angles allowed for this joint The first node that is parsed from the file is the head node of the hierarchy tree The program performs a search in the array of meshes to locate the mesh with the same name as the node and assigns its pointer to it After its centre min and max values are assigned the program checks if there are any children nodes keyword NODE instead of END In that case it allocates memory for the new node copies its pointer in the array of children nodes and repeats the same reading process for the new node until the END keyword is reached If a node has no children then it is a leaf node A sample part of the joint hierarchy file for the hierarchy of fig 4 2 is the following MESH man wrl NODE Hip centre 0 0 4 0 min 0 0 0 max 0 0 0 NODE Abdomen centre 0 0 4 0 min 1 45 0 6 0 5 max 0 45 0 6 0 5 NODE Chest centre 0 0 0 min 0 0 0 55 max 0 0 0 END Chest E ND Adbomen ODE Right_Thigh centre 0 036 0 35 0 01 min 0 0 0 max 0 0 0 NODE Right_Shin centre 0 037 0 2 0 01 min 0 0 0 max 0 0 0 END Right_Shin E ND Right_Thigh NODE Left_Thigh centre 0 036 0 35 0 01 min 0 0 0 max 0 0 0 NODE Left_Shin centre 0 037 0 2 0 01 min 0 0 0 max 0 0 0 END Left_Shin E Z D Left_Thigh
65. lso get a hint for the ball s position by the sound that it makes when it bumps to the floor The decide part is to choose if the virtual human should continue walking forward or rotate its body This decision is based on comparing the position of the ball with the 42 virtual human s walking direction The last phase is the action itself where the body will walk one step forward or rotate towards the ball More complicated dynamic actions involve catching a moving object or hitting it These actions must use a form of inverse kinematics because they involve more than one joint that have to be coordinated to succeed Instead of using a generic inverse kinematics solver a different approach is introduced This approach tests at every step the best rotation for each joint to achieve the target The set of joints or the joint chain that are going to be used by the system are first defined Then a function has to be assigned and return on how close to the target is the current state of the virtual human is The process is For each join in the chain For each degree of freedom of the joint increase and decrease angle by a value v check which of the two states improves the function if that state is different from the previous one decrease the value of v or else increase it if lower than the maximum speed assign the value to the angle This process is repeated in each frame and the system always corrects itself towards the
66. lve human activity e g fire in a crowded building and visualisation of intelligent agent systems are some target applications The growth of public interest in virtual environments distributed simulation and collaboration systems etc will increase the demand for real time simulated humans and many related application fields are about to appear in the near future 67 APPENDIX USER MANUAL 1 CD CONTENTS The CD that accompanies this dissertation has the following contents gt Program folder the executable program with all necessary data gt Library folder the additional OpenGL headers libraries and dynamic link libraries needed to compile and run the program in case they are not already installed gt Code folder the program code and workspace gt Document folder this dissertation in electronic format Word 97 All the files of the CD are explained in Table 1 Program Folder Simulated Human exe The executable program man dat woman dat The data files that can be loaded from the program man wrl woman wrl The VRML files that contain the mesh description of each body These files have been directly exported by Metacreations Poser pos files Files containing information joint angles for various key positions These are loaded by the program Library Folder bh files lib files These are the necessary OpenGL library files to compile the program in Microsoft
67. me systems and it has succeeded in being efficient even in environments with multiple virtual humans Its value is that it provides a programmer with all the critical design decisions that have been taken to reach this result a simulated human that utilises some of the ideas behind state of the art human modelling and simulation systems deformable skin collision detection and response inverse kinematics and is still efficient enough to be used in real time systems The use of object oriented 66 programming that ensures the project s reusability and extensibility is also of great importance There are many possible extensions of this project The fact that it does not have a predefined set of joints but can support any number of them allows one to extend it towards fields that require more detail such as grasping or facial animation All one has to do is to extract more complex models from Poser where the full set of finger joints and a lot of facial muscles are supported and then work on this specific field One other possible extension is to embed muscle forces that will lead to a more accurate physical simulation and then simulate actions like walking with forward dynamics rather than kinematics Numerous application areas could take advantage of simulated humans like the one presented in this dissertation Distributed virtual environments that might need more accurate and natural motion of avatars real time simulations that invo
68. nder of a mesh Wall A wall Ball A bouncing ball Scene The scene which contains a number of Person and Wall objects and a Ball Table 5 1 A short description of the classes used by the program 7 TESTING The testing of this application was not a very easy process This is due to the fact that the geometry files contain a very large polygon set and it is almost impossible to determine where an error has occurred Therefore the critical processes such as the VRML importer were built in small steps and each step was tested thoroughly before proceeding to the next one Each of the classes that are used to determine a mesh object Mesh Colour Face Point Vector has an output streaming function which is used for debug reasons A Face object for example displays the Colour and the set of Points that define it by using the streaming functions of the Colour and Point classes Whenever a new step in the VRML importer was completed e g the reading of vertices the memoty contents were streamed to an output file and compared with the input data When the VRML importer was ready global information about each mesh was exported and tested One sample part of an output file of that stage is the following Hip Vertices 226 Colours 10 Faces 213 Bounding box 0 0643671 0 413679 0 0404563 gt 0 0667226 0 320257 0 0529132 Abdomen Vertices 278 Colours 10 Faces 296 Bounding box 0 0602225 0 489162 0 04
69. ng marker swapping missing or noisy data and false reflections Tracking can be an interactive and time consuming process depending on the quality of the captured data and the fidelity required For straightforward data tracking can take anywhere from one to two minutes per captured second of data at 120 Hz For complicated or noisy data or when the tracked data is expected to be used as is tracking time can climb to 15 to 30 minutes per captured second even with expert users First time users of tracking software can encounter even higher tracking times Digital Poseable Mannequins 12 The Digital Image Design Monkey bttp www didi com www areas products monkey2 is the first commercial device to provide motion capture using poseable humanoid figures or other morphological types This class of devices should allow easier inclusion of those familiar with the stop motion work flow into computer animation While these devices may provide continuous data it is not usually continuous motion that is being captured but the position of the mannequin at discrete points in time The monkey consists of a poseable mannequin typically sub scale position encoders for each of the mannequin s joints and an electronic control unit which is attached to a host computer As the joints of the monkey are moved the control unit tracks their positions and passes them to a driver program on the host computer this may either be on r
70. nimised with the use of smaller timesteps and better integration methods but these have computational costs The primary aim of real time simulations including this project is not to be 100 accurate but to achieve a good balance between execution speed and realism Therefore in most of these cases the Euler Integration 36 method is being used which is the simplest form of integration but can still have realistic results if the timestep is small enough The physical simulation is conducted in discrete timesteps Each object has its own mass position and velocity and in each timestep its position and velocity are recalculated following the laws of kinematics and collision In this project the system executes the following commands in each timestep dr For each object 1 Calculate the total force F applied to it This will give rise to acceleration a such that F m a From this equation the system can find the value of a 2 If the velocity in the previous timestep was vp then the current velocity v is v v a dt 3 If the object s position in the previous timestep was D its new position p will be the D D D 0 p result of the mean velocity between the previous and the current timestep So v p p gt gt de 4 Check if the object collides with any other object in the scene If not assign the new position p and the new velocity v to the object and proceed to the next one 5 Assuming that the other obje
71. nslatef px py pz translation ender a function that renders the world LPopMatrix glPushMatrix glTranslatef 0 0 0 0 dist glRotatef 0x 1 0 0 0 0 0 rotation around x axis glRotatef Oy 0 0 1 0 0 0 rotation around y axis glRotatef 0z 0 0 0 0 1 0 rotation around z axis g E gl The program has four different camera states that can be selected from the menu and uses both camera models to implement them 3 VISUAL C AND MEC The project has been programmed in C There are two reasons for choosing C as an implementation language First of all the program had to be written in an object oriented language because its aim is not only to be a standalone application but also to have an easily extensible and reusable code Therefore languages like C or Pascal which does not have full object support are not suitable On the other hand the focus of this project is real time display capabilities therefore the program had to be as efficient as possible Java might be a good object oriented language but it is not as efficient as C because it uses an interpreter and it does not have OpenGL support As a conclusion the most efficient object oriented approach was C C alone is not enough to create graphical applications with OpenGL The programmer also needs a windows interface that will enable the application to run in Microsoft Windows This project is using the Microsoft Foun
72. nsure both quality and efficiency of the visual output The reason for choosing OpenGL 45 instead of any other 3D graphics library is because OpenGL takes full advantage of the hardware acceleration of current graphics cards and is therefore much mote efficient that any other software only approach The only other API which also uses graphics accelerators is Microsoft s Direct 3D The major drawback in using Direct 3D is that compared to OpenGL it is a much more low level library and one needs more time and effort to create graphics applications Additionally Direct 3D code can only work in PC s so one cannot create platform independent code and have his her applications running in powerful workstations 2 1 Polygon Rendering OpenGL does not only support triangles it can render any polygons This proved to be very important because the imported models were not triangle based so no additional triangulation method was needed A sample OpenGL code to display a polygon on screen is the following glColor3f r g biz assign colour glBegin GL_POLYGON for i 0 i lt n i for each edge glNormal13f normal i x normal i y normal i z glVertex3f vertex i x vertex i y vertex i z normal vertex r r glEnd This code displays a polygon with n edges and colour r g b where its normals and vertices are stored in normal and vertex arrays The rendering of the scene in the pro
73. nts and muscles able to generate forces Additionally there has to be an embedded behavioural model to help the body maintain its balance and move in an effective way There has been a great amount of research in the field of human motion and simulation and a number of different systems and approaches have been proposed Each of them vaties in appearance function and autonomy according to the application field and the required detail and accuracy Norman Badler probably the most important researcher in this field proposes three separated stages for the development of a virtual human Bader 1993 gt body modelling the visualisation of the overall human body and the definition of the skeletal structure with proper joint motions gt spatial interaction the direct manipulation of the human joints or the creation of goal oriented motion based on constraints with the use of inverse kinematics algorithms gt behavioural control the ability to generate complex motion following a simple set of tules that determine the model s behaviour Obviously these three stages differ in levels of complexity and each one is based on the previous 2 BODY MODELLING The process of modelling a virtual human involves three different stages The most primitive one is the visualisation of the body which is the same process as modelling any other 3D object This stage may be the only one required for producing static images but it is impos
74. ody parts fig 6 2 These postures can be exported to or imported from a file Importing and exporting posture information allows the user to define his her own keyframes for the predefined actions of the human model The available keyframes are 61 gt rest pos the rest position of the model gt walk1 pos walk2 pos key positions for walking animation gt sidel pos side2 pos key positions for walking sideways gt bend pos key position for bending down Figure 6 2 A posture of the human body 1 3 Physically based modelling The user can start the animation with the Edit Start Stop animation command from the program menu The ball then obeys gravity and bounces around the scene It can collide with the floor the walls and the human body it is actually colliding with the bounding cylinders of the body fig 4 9 When one or more collisions are detected the ball bounces off and loses some part of its energy It is not perfectly elastic so it will rest on the ground after a while 1 4 Walking animation and other actions 62 The user can control the virtual human provided that the animation mode is enabled With the cursor keys the human model can walk and rotate to any direction More specifically it can walk forward or backwards rotate left or right or walk sideways In case it hits the ball while moving the speed of the appropriate body part will be approximated and included in the calculation of
75. onference Series pp 11 20 Addison Wesley Longman Edited by Alyn Rockwood ROEHL B 1999 Specification for a standard humanoid Online Available http ece uwaterloo ca h anim spec1 1 accessed 9 March 2000 UNUMA M ANJYO K and TAKEUCHI R 1995 Fourier Principles for Emotion based Human Figure Articulation Proceedings of SIGGRAPH 95 Computer Graphics Proceedings Annual Conference Series pp 91 96 August 1995 Los Angeles California Addison Wesley Edited by Robert Cook VINCE J 1995 Virtual Reality Systems Addison Wesley Press WATT A and WATT M 1992 Advanced Animation and Rendering Techniques Theory and Practice Addison Wesley Press WELMAN C 1993 Inverse Kinematics and Geometric Constraints for Articulated Figure Manipulation MSc Thesis Simon Fraser University 73 WITKIN A and KASS M 1988 Spacetime Constraints Computer Graphics Proceedings of SIGGRAPH 88 vol 22 pp 159 168 WU Y KALRA P and MAGNETAT THALMANN N 1997 Physically based Wrinkle Simulation gt Skin Rendering EGCAS97 Budapest Hungary ZORDAN V B and HODGINS J K 1999 Tracking and Modifying Upper body Human Motion Data with Dynamic Simulation Computer Animation and Simulation 99 Eurographics Animation Workshop 74 ADDITIONAL BIBLIOGRAPHY AMAYA K BRUDERLIN A and CALVER T 1996 Emotion from Motion Graphics Interface 96 pp 222 229 Canadian Human Computer Communications Socie
76. orrespond to so the collision detection has to be done with the transformed cylinders and not with the initial ones Although this should complicate the calculations there is a trick that can avoid this complication Let us for example check if a transformed cylinder collides with a sphere Instead of transforming the cylinder one could inverse transform the position and velocity of the sphere and check for collision with the initial cylinder If the objects collided the new position and velocity of the sphere has to be transformed back to the original coordinate system The steps are 1 Find the matrix of the corresponding mesh and calculate its inverse 40 2 Multiply the inverse matrix with the centre and velocity of the sphere 3 Check if the sphere collides with the initial cylinder 4 If they do calculate the new position and velocity values and multiply them with the transformation matrix The collision detection of a primitive with a cylinder is not a very complex task Let us consider the case of an x aligned horizontal cylinder and test if a point is inside it Let the centre of the cylinder be cx cy cz its radius be R and its length Let also the point be px py pz The point is inside if and only if l l gt cx S lt px lt cx and 2 2 gt the distance of the point from the cylinder s axis is less than the radius In other words Vey px cz pz SR Similarly one can check if an edge
77. pports four different view modes Two of them have a different focal point the third one is fixed the user cannot rotate it while the fourth one displays additional primitives for testing and debugging reasons The view models are gt Standard This is the simplest mode The camera always focuses on the centre of the wotld gt Trace The camera focuses on the centre of the virtual human This mode is very useful for examining a model It is also valuable for tracing the model when in animation mode because it constantly follows its motion gt Eye The camera is fixed on the eyes of the virtual human so the user sees from the model s point of view This view is useful for virtual environments if the virtual human is used as an avatar gt Debug Same as the standard mode but additionally displays the bounding cylinders superimposed on the body parts 1 2 Loading models and defining postures The user can load human models on the program with the use of the File Open command from the menu The supported files are the joint hierarchy files dat which contain the joint tree information and the name of the VRML file wel of the body mesh Unless the name is in full path both files have to be in the same directory When the model is loaded it is displayed in the centre of the scene The user can then select any body part of the model and rotate it using the cursor keys Postures can be defined by rotating multiple b
78. rated by varying the local rotations applied at each joint over time as well as the global translation applied at the joint root There are two fundamental approaches the low level forward kinematics approach and the more elegant Anere kinematics one Forward kinematics involves explicitly setting the position and orientation of objects at specific frame times For skeletons this means directly setting the rotations at selected joints and possibly the global translation applied to the root joint to create a pose Using forward kinematics the position of any object within a skeleton can only be indirectly controlled by specifying rotations at the joints between the root and the object itself In contrast inverse kinematics techniques provide direct control over the placement of an end effector object at the end of a kinematic chain of joints providing a solution for the joint rotations which place the object at the desired location Welman 1993 Inverse kinematics sometimes called goal directed motion offers an attractive alternative to explicitly rotating individual joints within a skeleton An animator can instead directly specify the position of an end effector while the system automatically computes the joint angles needed to place the part Watt 1992 The inverse kinematic problem has been studied extensively in the robotics field although it is only fairly recently that the techniques have been adopted for computer animation
79. rithms for group behaviours obstacle avoidance and path following 3 5 Hybrid Approaches Besides Keyframing Motion Capture and Simulation there are also approaches that use a combination of these In most of the cases motion captured data are combined with simulation algorithms to retain the important characteristics of dynamic simulation while extracting the stylistic details found in human motion data Z Popovic and A Witkin propose an algorithm for transforming character animation sequences that preserves essential physical properties of the motion Popovic 1999 The algorithm constructs a simplified character model and fits the motion of the simplified model to the captured motion data From this fitted motion it obtains a physical spacetime optimisation solution that includes the body s mass properties pose footprint constraints and muscles From this altered parameterisation it computes a transformed motion sequence The motion change of the simplified model is mapped back onto the original motion to produce a final animation sequence This algorithm is well suited for the reuse of highly 17 detailed captured motion animations A similar approach Zordan 1999 is using simulation to animate models with different dynamic and kinematic parameters from motion captured data 4 APPLICATION FIELDS Virtual humans can be useful in many different application areas especially in cases where human activity has to be visualised
80. s For New Characters SIGGRAPH 97 Proceedings Los Angeles CA HODGINS J O BRIEN J and TUMBLIN J 1997 Do Geometric Models Affect Judgments of Human Motion Graphics Interface 97 pp 17 25 May 1997 Canadian Human Computer Communications Society Edited by Wayne A Davis and Marilyn Mantei and R Victor Klassen HODGINS J O BRIEN J and TUMBLIN J 1998 Perception of Human Motion With Different Geometric Models IEEE Transactions on Visualization and Computer Graphics 4 4 pp 307 316 ISSN 1077 2626 HODGINS J WOOTEN W BROGAN D and O BRIEN J 1995 Aximating Human Athletics Proceedings of SIGGRAPH 95 Computer Graphics Proceedings Annual Conference Series pp 71 78 Addison Wesley Edited by Robert Cook IWATA K MORIWAKL T and KAWANO T 1980 Computer graphics applied to human motion analysis and body force evaluation Eurographics 80 pp 167 178 North Holland Edited by C E Vandoni JUNG M NOMA T and BADLER N 1993 Collision Avoidance Planning for Task Level Human Motion Based on a Qualitative Motion Model First Pacific Conference on Computer Graphics and Applications Korean Information Science Society Korean Computer Graphics Society KOREIN J and BADLER N 1982 Techniques for Generating the Goal Directed Motion of Articulated Structures IEEE Computer Graphics and Applications 2 9 p 71 74 76 81 LAMOURET A and GASCUEL M 1995 Scripting Interactive Physically Bas
81. s and Dialogs Keyboard handling follows the same idea There are two functions OnKeyDown and OnKeyUp that are called when the user presses or releases a key respectively There are keys that activate actions of the virtual human and need to be pressed once These are handled directly in the OnKeyDown function On the other hand the cursor keys have a different role They make the virtual human walk as long as they are pressed These keys indicate that the walking process started in the OnKeyDown function and that it stopped in the OnKeyUp one When the program updates the screen it checks if the walking state is enabled or not and calls the appropriate method The menu commands are simpler to control since MFC provides a function for each menu command that can be called by the user The program also uses one dialog fig 5 1 which allows the user to select a body part of the virtual human and rotate it The 49 dialog contains a combo box that has all the available limbs and when the user presses OK it marks the pointer of the object that he she has selected Select Body Part Cancel Left Forearm Left_ Shoulder Figure 5 1 The dialog to select body part 33 Timer The application uses a global timer to generate the animation sequences There is a timer function provided by the MFC which is called at a constant timestep The length of this timestep defines the frame rate of the application Nevertheless
82. s but no size they can however be employed to describe the motion of very large objects without introducing significant errors Newton s first law of motion states that the momentum of a particle is constant when there are no external forces In other words an object moves in a straight line unless disturbed by some force In reality the universe is full of gravitational electric magnetic or atomic force fields and in general it is only when objects move away from the earth that they begin to obey this law Newton s second law of motion states that a particle of mass m subjected to a force F moves with acceleration F m This law is also called the equation of motion of a particle and means that work has to be done whenever an object is accelerated Newton s third law of motion states that f a particle exerts a force on a second particle the second particle exerts an equal reactive force in the opposite direction This law is useful when considering collisions between objects One cannot just code Newton s laws in a computer program and expect to have a perfect simulation There is a great difference between the theory and the implementation in computer programs in theory the time is considered to be continuous while computers work with discrete timesteps The numerical integration used in computer programs causes inevitable errors usually in the form of energy discrepancy A general rule is that the error can be mi
83. s to have fixed primitives usually spheres on the joints The major drawback of this approach is the fact that these primitives do not always fit perfectly with the model s meshes thus distorting the appearance of the model Another problem is that these spheres add a lot more polygons to the scene and decrease the performance significantly In this project a different approach has been adopted Instead of having primitives at the joints the idea is to have the ends of adjacent segments always connected with each other The human model in its initial pose has a continuous skin This means that adjacent meshes share common vertices and each one uses its own copy If the program can ensure that whenever a segment is rotated the vertices that are common with any adjacent segment will not be rotated then no crack is going to appear The effect of this method can be seen in fig 4 5 common vertices Figure 4 5 Segment rotation without and with common vertices In the beginning of the program when the model is loaded each mesh checks its adjacent ones if they share common vertices i e if there are vertices with the same value Let us suppose that the mesh of the segment S and that of S share the vertices v vz v Each of the common vertices that belongs to Sj is being deleted and replaced with a pointer to the corresponding vertex of S This means that all the polygons of S that contain one or more of the common vertices will
84. sible to animate the body having only its 3D representation Therefore an essential stage is the modelling of the skeleton which will define the moving parts of the body and the type of motion that they can perform These two stages alone are enough for simple animated humans Nevertheless the production of natural looking believable motion requires much more Both the skin and the clothes of the virtual human should move and deform in a natural way The calculation of the skin and cloth motion is the most computationally intensive task which is why it is not suitable for real time animation 2 1 Modelling the body Modelling the human body follows the same process as modelling any other 3D object There are many ways to represent a figure geometrically curved surfaces voxels etc but the most efficient one especially for real time applications is to model the body as a mesh a set of vertices and polygons The level of detail and complexity of the body strictly depends on the application area Games and real time simulations tend to use models with a small number of polygons and use alternative methods e g textures to display detailed parts On the other hand in video or high quality image productions the body is modelled with the maximum possible detail There are some commercial programs that allow users to define and construct a human body model and export it in a 3D file format probably the best known is Poser from Metacrea
85. simple walking animation may need less than a dozen joints On the other hand a virtual human that can grasp various objects requires a lot of additional joints and segments for the fingers of each hand In this project a virtual human can have an arbitrary number of joints and segments and any possible hierarchy tree to connect them This gives the user programmer the ability to load any possible models and adjust the program to the needs of different applications having any possible detail The human body representation is something that should also be arbitrary Therefore the program does not work with a fixed model but is able to load models dynamically The model should follow a file format and be a set of polygon meshes as described in the previous paragraph Additionally each mesh should have a unique name The process of loading a human body model is as follows for each mesh read the set of vertices read the set of normals and assign them to the vertices for each polygon find the corresponding vertices and assign pointers to them read the set of colours and assign them to the polygons Although the human body is now split into a set of independent polygon meshes they themselves are not enough to animate it The skeletal information is necessary to define how these meshes are going to be transformed This is loaded from a supplementary file which has the following data gt the joint hierarchy The tree structure t
86. t it cannot be applied to all body meshes The body has to be constructed using contours lying on planes perpendicular to the skeleton s segments In cases where a high degree of realism is required the skin s motion has to follow the laws of physics it has to behave as an elastic surface One way to achieve this is to model the skin as a set of springs and masses Vince 1995 Each vertex of the surface mesh can be treated as a small mass and each edge as a spring with a relatively high restitution factor This structure will ensure the elastic deformation of the skin while the body is moving It is nevertheless required that the surface mesh is as regular as possible and it should be relatively dense as well There are also more complex approaches that use biomechanical models to determine the skin s behaviour and generate wrinkles while the skin is deforming Wu 1998 7 Magnetat Thalmann 1996 These models additionally take the muscles of the body into consideration and also calculate their deformation Cloth simulation follows almost the same process using a spring and mass model again The sutface of the clothes has to be a dense set of small masses which should have gravity and be able to collide with themselves and the skin underneath A more detailed model large set of springs and masses will result in a more realistic deformation and wrinkle generation of the clothes Today s computational power allows real time clot
87. taches the ball to the man s hand the C key causes a kick and the B key makes the man follow the ball until it catches it and throws it back to the ground The demo mode can be initialised from the Edit Demo command of the menu All available keys and menu commands of the program are summarised in the following paragraph 3 SUMMARY OF KEYS AND MENU COMMANDS Key Description Q W Positive negative rotation of the active body part around the x axis A S Positive negative rotation of the active body part around the y axis ZAX Positive negative rotation of the active body part around the z axis Up Down arrow human walks forward backwards Left Right arrow human turns left right SHIFT Left human walks sideways to the left right Right arrow C human performs a kick SPACE human grasps the ball B human tries to catch the ball and throws it back Mouse Description left button rotates the camera mouse move tight button Zooms camera in or out mouse move up or down Menu Command Description File Open loads a human model File Exit exits the program Edit Change Changes the active body part Active Edit Export Saves the current posture to the posture dat file 70 Postute Edit Import loads a posture from the posture dat file Posture Edit Start Stop starts stops the human and b
88. tems are based on high contrast video imaging of retro reflective markers which ate attached to the object whose motion is being recorded The markers are small spheres covered in reflective material such as Scotch Brite The markers are imaged by high speed digital cameras The number of cameras used depends on the type of motion capture Facial motion capture usually uses one camera sometimes two Full body motion capture may use four to six or more cameras to provide full coverage of the active area To enhance contrast each camera is equipped with infrared IR emitting LEDs and IR pass filters are placed over the camera lens The cameras are attached to controller cards typically in a PC chassis Depending on the system either high contrast 1 bit video or the marker image centroids are recorded on the PC host during motion capture Before motion capture begins a calibration frame a carefully measured and constructed 3D array of markers is recorded This defines the frame of reference for the motion capture session After a motion capture session the recorded motion data must be post processed or tracked The centroids of the marker images either computed then or recalled from disk are matched in images from pairs of cameras using a triangulation approach to compute the marker positions in 3D space Each market s position from frame to frame is then identified Several problems can occur in the tracking process includi
89. the animation the program calculates and stores the difference between the initial and final rotation for each joint D R R The animation is split in frames and each frame has a time difference d from its previous one Interpolating between the two poses means simply adding to each joint rotation a small fragment of the difference The process is For each frame k repeat 33 For each joint i R R dD until k dt t This generates a smooth transition between two states poses The use of this basic process is made clear in the following example the walking animation 4 4 Walking animation The animation of a walking human is a typical keyframing example This is because of the nature of walking which is a continuous transition between several states The aim of walking is not only to continuously change the joint rotations but also the global translation of the body causing the human to move towards a certain target Walking has the following states gt State 1 the left leg touches the ground and pushes the body forward gt State 2 the right leg is on the ground and drives the body Nevertheless a walking animation should contain two more states One is the transition from current position to state 1 to start walking and the other is the transition from any state to a rest position to stop walking During each state the virtual human is also changing its global translation to animate the body motion This
90. the collision response Besides walking the program also supports other actions such as kicking the ball grasping the ball or following the ball and trying to catch it The last action is the most complex one and implements the inverse kinematics approach discussed in the Design chapter The virtual human is constantly walking towards the ball following its motion When it gets close to it it stretches its arm and tries to catch it if the ball is resting on the ground 1t bends down to approach it better fig 6 3 In the process of catching the ball the forearm shoulder and abdomen are constantly rotated until the end effector Le the arm reaches the ball Figure 6 3 The man tries to catch the ball 1 5 Demonstration The program offers a demonstration option where all the features of the virtual human are presented The program loads one male and one female model Their initial positions are on opposite sides of the room When the animation starts one model tries to catch the ball When it succeeds it turns to the other one and throws the ball to it The other model is now trying to catch the ball while the first one is returning to its initial position This sequence continues infinitely and the user is able to rotate the scene and switch 63 between the cameras The most suitable camera mode is the trace mode which is always focusing on the person that is chasing the ball Figure 6 4 A sample screenshot from the
91. the shadowed object s surface outlined by the object casting the shadow appears dark A shadow can be produced by flattening the original object into the plane of the surface which in this case is the ground plane The idea is to create a transformation matrix that flattens any 3D object into a two dimensional polygon No matter how the object is oriented it is squashed into the plane in which the shadow lies This matrix depends on the position of the light source and the coefficients that determine the plane 3 MODELLING THE BODY A virtual human is a part of the scene therefore it has to be represented in terms of polygons like any other object The human body could be one big polygon mesh but it would be very difficult to animate it If for example the virtual human had to raise its arm the program would have to find which vertices and polygons belong to the arm and rotate only these A much better approach is to define the body as a set of smaller meshes each one representing a segment limb Each segment is the body part that connects two joints and should maintain its geometry throughout the animation 25 The number of joints and segments that will be defined on the human body depends on the application and the animation detail required So does the number of degrees of freedom allowed per joint A real human has in total over two hundred degrees of freedom but efficient animation can be produced with significantly less A
92. the use and structure of the joint hierarchy files The program s classes and the testing methods that have been employed are also described 2 OPEN GL OpenGL is a library of graphics routines available on a wide variety of hardware platforms and operating systems including Windows 95 Windows NT OS 2 Digital s Open VMS and X Windows It was developed by Silicon Graphics Incorporated in the year 1992 and was eventually accepted as an industrial standard for hardcore 3D graphics The OpenGL Architecture Review Board maintains the library ensuring that OpenGL is implemented properly on various platforms This makes porting the programs across the platforms much easier OpenGL routines are well structured highly stable intuitive scalable from PCs to Super Computers and guaranteed to produce consistent visual displays across various platforms OpenGL is defined as a software interface to graphics hardware In essence it is a 3D graphics and modelling library that is extremely portable and very fast OpenGL is not a programming language like C or C It is more like the C runtime library which provides some pre packaged functionality There really is no such thing as an OpenGL program but rather a program the developer wrote that uses OpenGL as one of its APIs There has been extensive use of the OpenGL graphics library in this project The camera control and rendering of the 3D scene is based on OpenGL commands which e
93. tices for smooth display The n vector in the vector array is the normal vector of the n vertex in the point atray gt the color array in the Color node it has all the colours that are used in the mesh each polygon can have a different colour The colours ate in RGB format ranging from 0 0 to 1 0 for each of the red green and blue values gt the colorIndex array it has the indices of the colours for each polygon For example if the n number in the colorIndex array is 1 it means that the n polygon of the mesh should have the second colour from the color array The VRML parser locates all Shape nodes in the file and for each one of them it reads all the previously stated arrays and stores the values in memory point array The program continuously reads strings from the file using the standard C streaming operators until it locates the point keyword It reads another string the f character and then starts reading vectors until a character is read The number of points in a mesh is not pre defined therefore the use of a static array is not suitable because it would have great memory consumption On the other hand the use of double linked lists would slow down the execution of the program since tracking the n element of a list needs at least n steps The way to get around this was to use a dynamic list for intermediate storage and then when the reading is over and the number of verti
94. tions hi p www curiouslabs com With such programs artists are able to select a primitive model from a library specify a number of visual details such as height muscle 4 size face characteristics and use this model in their own programs or animation sequences Fig 2 1 Figure 2 1 Human figure by Metacreations Poser 2 2 Modelling the skeleton Proper motion requires a skeleton to be represented underneath the skin of a human body to define the moving parts of the figure Although it is possible to model each of the bones in the human body and encode in the model how they move relative to each other for most types of motion it is sufficient to model the body segments in terms of their lengths and dimensions and the jo nts in terms of simple rotations The joints are usually rotational but may also be sliding Each rotary joint may allow rotation in one two or three orthogonal directions these are the degrees of freedom of the joint A detailed approximation to the human skeleton may have as many as 200 degrees of freedom although often fewer suffice Restrictions on the allowable range of movements for a joint can be approximated by limiting the rotation angle in each of the rotation directions at each joint The individual objects comprising the skeleton are each defined in their own local coordinate systems and are assembled into a recognisable figure in a global world coordinate system by a nested series of
95. to each vertex and stored together in the mesh structure The scene structure of this project is depicted in the following class diagram Fig 4 1 Vector Figure 4 1 The class diagram of the scene elements 24 As seen in the diagram each mesh has an arbitrary number of points and an arbitrary number of polygons Each polygon can point to three or more points and each point has two vectors the vertex and the normal Besides these a material colour is associated to each polygon and the scene can also have an arbitrary number of Zeit Each light is considered to be a pointlight having a unique position in space and emitting equal energy in all directions The final image that will be rendered from the scene depends on the position and orientation of the camera Three dimensional geometry and shading can give a relatively good perception of the third dimension Nevertheless the exact position of objects in the virtual space is better perceived with the use of shadows Therefore one additional feature of this project is the rendering of shadows cast by the objects onto the ground This is especially important in animation sequences where the distance of the virtual human s feet to the ground can be visually estimated Conceptually drawing a shadow is a simple process A shadow is produced when an object blocks light from a light source from striking another object or surface behind the object casting the shadow The area on
96. ty Edited by Wayne A Davis and Richard Bartels BADLER N CHI D and CHOPRA S 1999 Virtual human animation based on movement observation and cognitive behavior models Computer Animation Conference Geneva Switzerland BADLER N MANOOCHEHRI K and WALTERS G 1987 Articulated Figure Positioning by Multiple Constraints IEEE Computer Graphics and Applications 7 6 pp 28 38 BADLER N O ROURKE J and KAUFMAN G 1980 Special problems in human movement simulation Computer Graphics Proceedings of SIGGRAPH 80 74 3 pp 189 197 BEZ H E BRICIS A M and ASCOUGH J 1996 A collision detection method with applications in CAD systems for the apparel industry Computer aided Design 28 1 pp 27 32 Elsevier Science CALVERT T CHAPMAN J and PATLA A 1982 Aspects of the Kinematic Simulation of Human Movement TEEE Computer Graphics amp Applications 2 pp 41 48 CALVERT T CHAPMAN J and PATLA A 1982 The simulation of human movement Graphics Interface 82 pp 227 234 CRAWFORD M 1977 Interactive computer graphics for simulation displays Proc Haman Factors Society 21 Annual Meeting pp 14 17 75 FARENC N BOULIC R and THALMANN D 1999 An Informed Environment Dedicated to the Simulation of Virtual Humans in Urban Context Computer Graphics Forum 183 pp 309 318 September 1999 Blackwell Publishers HODGINS J and POLLARD N 1997 Adapting Simulated Behavior
97. ually the multiplication of its vectors by a matrix Both vertices and normal vectors should be multiplied by a special 3x3 matrix the rotation matrix The following matrix rotates points about an arbitrary axis passing through the origin 29 2 ko c ka a sa ka a 8a Z x gt 2 ka a sa ka c ka a sa 2 ka a sa ka a sa ka c where a a a is the unit vector aligned with the axis O is the angle of rotation c os s sm and k 1 cos To rotate one point around the three angles rx ry rz one should generate three rotation matrices and multiply them together The three matrices are gt R with a a a 100 and 0 rx gt R with a a a 010 and O ty and gt R with a a a 001 and 0 rz For all vertices v v v v and normals n n n n v RR R v R v and n R n where vi and n are the vertices and normals after the rotation and R is the 3x3 matrix that results from the product of R R and R This process rotates a mesh around the world centre 0 0 0 which is not always the desired case To rotate the segment S around the joint centre P one should For each vertex v and normal n subtract the centre of rotation from the vertices v v P rotate the vertices and normals v Bal n Rn add the centre of rotation to the new vertices v v P H This is still not sufficient The rotation of the

Simulated Human being a dissertation submitted in partial fulfilment

Contents

Download Pdf Manuals

Related Search

Related Contents