Crowdbrush: Interactive Authoring of Real-time
Contents
1. 1 3 3 Picture Image Generation Display algorithms 1 3 4 Graphics Utilities Graphics editors I 3 7 Three Dimensional Graphics and Realism Animation 1 Introduction With current consumer grade personal computers it is possi ble to display 3D virtual scenes with thousands of animated individual entities at interactive framerates using different techniques as animated impostors TLC02 or point based rendering WS02 When increasing the number of involved individuals it is becoming more difficult to create unique and varied content of scenarios with large numbers of entities If we want to create or modify features of every single individual one by one it will soon become too laborious If on the other hand we apply a set of features either uniform or patterned to many individuals at once it could create unwanted artefacts on a larger scale resulting in an army like appearance with too uniform or periodic distributions of individuals or char acteristics Use of random distributions can alleviate such problems however it can be very difficult to capture the de sired constraints into a set of mathematical equations espe cially considering integration into common art production pipelines The Eurographics Association 2004 The challenge is how to create complex scenes resembling a variety rich look of the real world Here we understand complexity analogous to a notion of complexity of patterns generated by2. Spatial configuration is qualified by properties of posi tion and orientation Appearance is influenced by prop erties of colour texture material and scale Execution of an animation action is determined by animation selec tion shift and speed properties High level features can use a combination of several low level features accessed through their properties For example a particular emo tional state would set animations from a predefined set with some specific speed or clothing style would select a set of appropriate textures and colours for different body parts The set of values of all instance properties fully de fines the configuration of the crowd members and is used to save and restore the state of the scenario during and across the authoring sessions 4 Behavior A behavior is not the focus of this article Nevertheless as more complex scenarios require some sort of behavior sim ulation MTO1 we implemented a simple spatial displace ment of the humans and a rule based behavior engine UT02 to handle more autonomous characters We opted for a sim ple reactive rules system as we needed fast execution for up to several thousands of agents Behavior rules can react to both internal and external events to the agents triggering se quences of actions as displacement or animations Agents can move around the environment following paths which are sequences of places with potentially variable sizes The list of all places
3. mated characters and objects from several computer games such as four legged animals insects helicopters or a forklift The main limitation of our rendering approach is a rela tively large amount of memory required in case more tem plates and more animations are needed to increase the va riety of a crowd However this problem is common to all current methods for real time rendering of large crowds as impostors or point based rendering which have memory re quirements several times higher Figure 9 Crowd rendered with coloured spotlight Figure 10 Crowd rendered with specular highlights shad ing The Eurographics Association 2004 B Ulicny P de Heras Ciechomski amp D Thalmann Crowdbrush Interactive Authoring of Real time Crowd Scenes Figure 11 Crowd rendered with cartoon shading 8 Conclusions and Future Work We presented a novel technique for authoring real time crowd scenes in conjunction with a management of large real time rendered crowds The benefits of our approach are twofold an intuitive and familiar 2D interface WYSIWYG for authoring and inter acting with crowds of humans in real time and the ability to render a large number of fully skinned and animated 3D characters in real time in a standard way that is easily inte grated into an existing triangle processing pipeline such as OpenGL We tested our application on authoring the scenario of a virtual audience as seen in Figure
4. P de Heras Ciechomski amp D Thalmann Crowdbrush Interactive Authoring of Real time Crowd Scenes stances of the same template have the same mesh and skele ton One of the important considerations we had to take into account was the ease of data production using already ex isting 3D content production tools We wrote a mesh and skeleton exporter for 3ds max using MaxScript 3DS04 The mesh itself consists of vertices and triangles The ex ported triangles have associated information such as texture coordinates materials and skeleton bindings The next step in the data production pipeline is exporting a reference mesh and a reference posture describing the initial animation for applying skeletal transforms using the Bones Pro 3 plug in BP304 This allows exporting of any animation for a par ticular skeleton The animation is defined as a list of bones in the hierarchy with translation and rotation for every sampled keyframe For importing the model into the engine we have to first unpack it into complete meshes For every animation we define a running time in seconds and every second is sam pled thirty two times to preserve fluidity of animation Sam pling is done on skeletal keyframes using spherical interpo lation on rotation quaternions and using a linear interpola tion for the translations The sampled skeleton is applied to the mesh and deforms it using conventional skinning meth ods Lan99 When the deformed mesh has be
5. and paths is stored by the en vironment manager which responds to queries from a rule system Positions of the agents between waypoints are in terpolated taking into account their desired speed and actual simulation update rate Along the way agents play looping The Eurographics Association 2004 B Ulicny P de Heras Ciechomski amp D Thalmann Crowdbrush Interactive Authoring of Real time Crowd Scenes keyframed animations of different walking styles or running depending on the displacement speed The behavioral rules and the environment database are ed itable in run time using rules and environment editors con nected with the behavior engine We added also a simple collision avoidance system based on a force based model HM95 In order to improve effi ciency of the collision avoidance spatial queries need be op timized otherwise collision detection would become a per formance bottleneck much sooner than rendering We used a simple bin lattice space subdivision to minimize the number of collision queries Rey00 The brush metaphor is integrated with behavior simula tion in a direct mode by sending events to the rule system and the environment manager The spray can be then used for example to send events to activate behavior rules which send an agent to a path or to add environment objects used for navigation and behavior control The brushes can also be used to control the behaviors in a non direct mode where t
6. scene we could change their behaviors by us ing the emotion spray that makes the human play differ ent kinds of animations from predefined sets for example The Eurographics Association 2004 B Ulicny P de Heras Ciechomski amp D Thalmann Crowdbrush Interactive Authoring of Real time Crowd Scenes Rendering without lighting 350 5 1134 triangles per human 620 triangles per human 300 330 triangles per human 250 200 150 Frames per Second 100 50 4 0 T T T T 200 300 400 500 600 700 800 900 1000 1500 2000 Number of Humans Figure 8 Crowd following paths defined by path brush happy animations or sad animations The particular state of the audience can be saved and then restored see Section 3 so that for example different reactions to different actors are prepared and then replayed when needed Virtual pedestrians In the second scenario the tasks are given models of a virtual city and template humans to dis tribute pedestrians to desired locations in the city to set up network of paths among specified locations and finally to let selected pedestrians follow these paths In this scenario we used eight different template humans with around one and a half thousand triangles each the model of the city has around sixteen thousand triangles In comparison with the previous scenario the placement of the newly created humans was not restrict
7. selected area in the world space changes with the level of zoom into the 3D scene This provides an intuitive control of focus if we want to work on a large part of the crowd we zoom out of the 3D view to see a desired part or whole of the crowd If we want to focus on a smaller group or individual we zoom in Thus features that are observable from a particular zoom level are also editable at this level Operators define how selection will be affected For exam ple a stroke of the creation brush with the random opera tor would create a random mix of entities see Figure 3 a stroke of the uniform colour brush would set colours of affected individuals to the same value as shown in Figure 4 and a stroke of the orientation brush with the gradi ent operator would let individuals turn in the direction of the gradient see Figure 5 In the current implementation the distribution of values for random operator is uniform it could be extended to use different distributions for ex ample by drawing 2D distribution curves in screen space Instance properties are non sharable attributes giving eo 4 itl Coats tt 5 rata ed F ih oO Hi UMMA CEU IA Figure 5 Orientation brush with gradient operator Hugo model by Laurence Boissieux C INRIA 2003 uniqueness to every individual member of the crowd In stance properties encapsulate lower level features influ encing both appearance and animations of virtual humans
8. 1 and on authoring of a pedestrian crowd see Figure 8 We concluded that it was easy to configure the scenes according to our wishes Our initial tests with an artist user confirm benefits of the intuitive interface the artist was able to create crowd scenes using the CrowdBrush after a few minutes instruction with out the need of lengthy tutorial and detailed user manual The user was provided only with a one page quick reference card displaying keyboard shortcuts associated with different brushes There are several directions for future work We will con centrate on extending the possible number of handled enti ties and on improving behavior repertoire Our preliminary tests with level of details show a promising increase in per formance or conversely an increase in number of humans while keeping interactive framerates Also a billboarding ap proach for far away humans is being incorporated similar to the one in TLC02 and it will be interesting to see how these two will work in conjunction For the user interface brush parameters specification can be improved by having dedicated widgets allowing to mod ify the properties of the active brush For example the actual The Eurographics Association 2004 colour used by a colour brush could be selected from a stan dard colour palette dialog or a template for a creation brush could be selected from a swatch of available virtual humans 9 Acknowledgements We are grate
9. Eurographics ACM SIGGRAPH Symposium on Computer Animation 2004 R Boulic D K Pai Editors Crowdbrush Interactive Authoring of Real time Crowd Scenes Branislav Ulicny Pablo de Heras Ciechomski and Daniel Thalmann Virtual Reality Lab EPFL CH 1015 Lausanne Switzerland Abstract Recent advances in computer graphics techniques and increasing power of graphics hardware made it possible to display and animate large crowds in real time Most of the research efforts have been directed towards im proving rendering or behavior control the question how to author crowd scenes in an efficient way is usually not addressed We introduce a novel approach to create complex scenes involving thousands of animated individuals in a simple and intuitive way By employing a brush metaphor analogous to the tools used in image manipulation programs we can distribute modify and control crowd members in real time with immediate visual feedback We define concepts of operators and instance properties that allow to create and manage variety in populations of virtual humans An efficient technique allowing to render up to several thousands of fully three dimensional polyg onal characters with keyframed animations at interactive framerates is presented The potential of our approach is demonstrated by authoring a scenario of a virtual audience in a theater and a scenario of a pedestrian crowd in a city Categories and Subject Descriptors according to ACM CCS
10. ain limitation of our spatially oriented brush metaphor is a weak control of time related aspects of sce narios We address this issue by incorporating behavior rules engine that is complementary to brushes see Section 4 Fur thermore as the Crowdbrush is integrated into the VHD framework the other authoring methods provided by the platform as Python scripts can be used in conjunction with the brush metaphor The advantage of our rendering approach is that polygo nal models are fully integrated into the triangle processing pipeline allowing for lighting see Figure 9 or even more complex shading effects For example it was straightfor ward to allow the crowdbrush application to render a crowd with specular highlighting see Figure 10 or cartoon shad ing see Figure 11 effects using an existing library of ef fects OSG04 As well our rendering pipeline is using only standard OpenGL features thus being independent of any hardware specific extension Furthermore because all characters are fully animatable using a skeleton it would be possible to integrate our crowd with some more complex motion control methods beyond simple keyframes For example if the scenario would re quire it some particular members of the crowd could be come animated by procedural walk generation inverse kine matics or motion graphs Our rendering engine is flexible to go beyond humanoid characters We successfuly imported fully skinned and ani
11. an potentially have unique values and can be individually selectable Responsiveness It must be fast enough for real time edit ing to allow for an immediate feedback loop Therefore it must not involve lengthy preprocessing stages at least for features that are target for authoring and also the system s responses to changes must be fast The requirements are close to those of any interactive ap plication Indeed boundaries between authoring and inter action are getting more and more fuzzy A recent trend in game development is to use the same application for part of the authoring and actual gameplay or to integrate play test ability in authoring tools RWS04 GBO04 The player in teraction is then the subset of possible interactions for a de signer Such designs came from the necessity of constantly switching between play and create modes while devel oping the game in order to increase productivity Figure 2 shows an overview of the system design The user controls the application using a mouse and a keyboard The mouse moves the visual representation of the brush tool we used an icon of a spray can on the screen with the mouse buttons triggering different actions either on render ing or behavior subsystems The keyboard selects different brushes sets their parameters and switches between navi gate and paint modes In the navigate mode the mouse controls position and orientation of the camera In the paint m
12. cellular automata as in Wol02 not uniform not periodical nor just fully random Bottom up approaches such as local rule based flocking Rey87 can create such complexity however they are diffi cult to control if we want to achieve particular end configu rations how to set local rules to get a global result In the recent work Anderson et al AMC03 achieved interesting results for a particular case of constrained flocking anima tion Nevertheless the algorithm can get very costly when increasing the number of entities and simulation time Major 3D content creation packages used by the media in dustry now offer tools to improve working with a large num ber of virtual characters CS04 SIB04 The most advanced crowd animation system for non real time productions is Massive used to create battle scenes in the Lord of the Rings movie trilogy Mas03 The production of massively popu lated scenes is still in the majority of cases a lengthy and B Ulicny P de Heras Ciechomski amp D Thalmann Crowdbrush Interactive Authoring of Real time Crowd Scenes jai 4 HA f P i 4 33 a j ggas F pny d 1 j p K A A i Pu 4 A i 4 P 3 s 4 ey H zea d X i Af Se a IM i d A vo ie ve a m mri 4 w gan i i neki x a a 3 A We ee 5 A w m Ty oon 4 rk oe j J 4 y M i f F 5 i 4 gt t y n r f N H A TT j ie vog Ha Sill aa
13. ed to a grid cre ation brush could add pedestrians at any desired location in the 3D world The only constraint was that for a drag mo tion of the brush the new human had to be at least at some The Eurographics Association 2004 i Rendering with lighting enabled 250 1134 triangles per human 620 triangles per human 200 4 330 triangles per human o fa 150 T Q A 100 4 S L 50 4 0 i T T T T T T T T T T 1 200 300 400 500 600 700 800 900 Number of Humans 1000 1500 2000 minimal distance from the previously created one Even this constraint is not completely necessary as force based colli sion avoidance will take care of not letting humans overlap However we found that imposing it was giving more aes thetically pleasing results In addition to creation of humans we used the brush to add sequences of places to the environment Our environ ment model allows to have places with different sizes which allows for example to channel a flow of the crowd to wider or more narrow streets The paths can be of three types de pending on the desired behavior on the boundaries of the path At the end of a simple path the agent stops a cyclic path lets the agent again move towards the first waypoint and a teleport path sets the position of the agent to the first waypoint Changes of the place size and designation of paths from active set of places is in the current imple
14. en computed the result is stored into a display list SWND0O3 If a char acter has more than one material which in our case is more than one texture the character is divided into several display lists each with its own material This allows us to exchange textures for different parts of the body in run time Rendering Management For management of the scene we use the OpenSceneGraph 3D graphics toolkit OSG04 where the crowd renderer is integrated as a custom node us ing OpenGL rendering SWND0O3 After all characters have been loaded into memory they are kept in a database called the manager The manager stores template humans and also keeps record of all exist ing instances A template human consists of a set of anima tions which in turn keep lists of display lists and textures An instance keeps hold of the properties for the human giv ing uniqueness to every individual see Section 3 The in stance properties are animation index time inside anima tion speed of animation mesh colour mesh scale material settings if there are many possible for example different textures are possible for the torso face and so on and a transformation matrix for translation and rotation Before the start of rendering all initial instances of the template humans are registered and their properties are set When the rendering loop starts the manager is called to render all instances simultaneously Rendering is optimized to minimize
15. ful to Mireille Clavien and Rachel de Bondeli for the design of virtual humans and scenes This work has been supported by the Swiss National Research Foundation and the Federal Office for Education and Science in the framework of the European project ERATO References 3DS04 3ds max 2004 http www discreet com 3dsmax 6 AMC03 ANDERSON M MCDANIEL E CHENNEY S Constrained animation of flocks In Proc ACM SIGGRAPH Eurographics Sympo sium on Computer Animation SCA 03 2003 pp 286 297 1 Bau98 BAUDISCH P Don t click paint Using toggle maps to manipulate sets of toggle switches In Proc UIST 98 1998 pp 65 66 2 BP304 Bones Pro 3 2004 http www digimation com 6 CS04 Character Studio 2004 http www discreet com products cs 1 GBO04 Gamebryo game engine 2004 http www ndl com 3 HH90 HANRAHAN P HAEBERLI P E Direct WYSIWYG painting and texturing on 3D shapes In Proc SIGGRAPH 90 1990 pp 215 223 2 HM95 HELBING D MOLNAR P Social force model for pedestrian dynamics Phys Rev E 51 1995 4282 4286 5 KMM 02 KALNINS R D MARKOSIAN L MEIER B J KOWALSKI M A LEE J C DAVID SON P L WEBB M HUGHES J F FINKEL STEIN A WYSIWYG NPR Drawing strokes directly on 3D models In Proc SIGGRAPH 02 2002 pp 755 762 2 Lan99 LANDER J Over my dead polygonal body Game Developer Magazine May 1999 1 4 6 Mas03 Massive crowd ani
16. he desired agents are first tagged by spraying with a tagging brush Later this tagging information is used to selectively trigger different behavioral rules for different agents allow ing for multiple actions to happen in different parts of the scene at the same time 5 Rendering When designing our rendering system we set out to achieve several goals following the requirements introduced in Sec tion 2 The first condition was to be able to render and manage simultaneously at least one thousand humans while keeping an interactive framerate Second we needed a va riety of appearance and behavior in the crowd while con straining memory usage We should also be able to render the humans with lighting In order to keep the data pipeline from model to rendering engine smooth we needed fast and robust pre processing of data Existing Approaches We considered several already ex isting crowd rendering approaches Tecchia et al TLC02 used billboards or impostors for crowd rendering The main advantage is that impostors are very light weight to ren der once they are in the memory of the graphics card The method requires building of all animations from all possi ble camera angles and storing these pictures in a texture One such texture can hold one frame of animation in very low resolution billboards where every individual sub frame is about 16 pixels tall and wide This can give good visuals since it is basically an image based rende
17. he third alternative was to use vertex shaders to deform a conventional mesh using a skeleton The disadvantage would be that the pipeline would be constrained to shaders and ev ery interaction such as lighting shadows and other standard effects would then have to be programmed with shaders The method of vertex tweening was used in the game Quake Qua96 for a smaller number of virtual humans In this approach fully precomputed meshes are exported by the 3D designing software since dynamic skinning was not fea sible at the time of game release The main disadvantage of this approach is complex and tedious data preparation and exporting pipeline Our main idea is to store complete meshes in OpenGL display lists and then carefully sorting them taking cache coherency into account while rendering This results in a method which has little or no processing on the CPU be cause precomputed meshes are stored on the graphics card The only consideration we had to do was to manage the ren dering so that humans that have the same materials meshes or animations are rendered at the same time since graphics cards need to be kept cache coherent to perform well Figure 6 shows an overview of our rendering pipeline Data Processing We chose to use skeletal models with deformed meshes as the basis of our animation system Our crowd consists of a large number of individuals that are cloned from a smaller number of template humans All in B Ulicny
18. http www renderware com renderwarestudio html 2 SALESIN D BARZEL R Adjustable tools An object oriented interaction metaphor ACM Transactions on Graphics 12 1 1993 103 107 2 Softimage XSI Behavior 2004 http www softimage com products behavior 1 SHREINER D WOO M NEIDER J DAVIS T OpenGL Programming Guide The Official Guide to Learning OpenGL Version 1 4 Ad dison Wesley 2003 6 TECCHIA F LOscos C CHRYSANTHOU Y Image based crowd rendering IEEE Computer Graphics and Applications 22 2 March April 2002 36 43 1 5 9 TOBITA H VelvetPath layout design system with sketch and paint manipulations In Proc Eurographics 03 Short Presentations 2003 2 UT02 Wol02 WS02 XLTPO3 ZHH96 ULICNY B THALMANN D Towards interac tive real time crowd behavior simulation Com puter Graphics Forum 21 4 Dec 2002 767 775 4 WOLFRAM S A New Kind of Science Wol fram Media Inc 2002 1 WAND M STRASSER W Miulti resolution rendering of complex animated scenes Com puter Graphics Forum 21 3 2002 Proc Eu rographics 02 1 5 XU S LAU F C M TANG F PAN Y Advanced design for a realistic virtual brush Computer Graphics Forum 22 3 2003 533 542 Proc Eurographics 03 2 ZELEZNIK R C HERNDON K P HUGHES J F SKETCH An interface for sketching 3D scenes In Proc SIGGRAPH 96 1996 pp 163 170 2 The Eurographics Associati
19. mation software for visual effects 2003 http www massivesoftware com MAY04 Maya 2004 http www alias com maya 2 MTO1 OSG04 PHL 98 PPM 02 Qua96 Rey87 Rey00 RWS04 SB93 SIB04 SWNDO3 TLCO2 Tob03 B Ulicny P de Heras Ciechomski amp D Thalmann Crowdbrush Interactive Authoring of Real time Crowd Scenes MUSSE S R THALMANN D A hierarchi cal model for real time simulation of virtual hu man crowds IEEE Transactions on Visualiza tion and Computer Graphics 7 2 April June 2001 152 164 4 OpenSceneGraph 2004 http www openscenegraph org 6 8 PIGHIN F HECKER J LISCHINSKI D SZELISKI R SALESIN D H Synthesizing realistic facial expressions from photographs In Proc SIGGRAPH 98 1998 2 PONDER M PAPAGIANNAKIS G MOLET T MAGNENAT THALMANN N THALMANN D VHD real time development framework architecture Building flexible and extendible VR AR systems with reusable components In Proc Computer Graphics International 2002 2002 3 Quake game homepage 1996 http www idsoftware com games quake quake 3 REYNOLDS C W Flocks herds and schools A distributed behavioral model In Proc SIG GRAPH 87 1987 pp 25 34 1 REYNOLDS C W Interaction with groups of autonomous characters In Proc Game Devel oppers Conference 00 2000 pp 449 460 5 RenderWare Studio game development plat form 2004
20. mentation done via keyboard Once the paths were defined we could send events to se lected pedestrians by spraying them with the event brush The crowd behavior system then responds to received events and let the agents move following the paths Furthermore we can affect speed of movement of selected agents by spraying them with the event brush now sending a different type of event Then the event triggers a rule which changes speed of locomotion and the associated animation for example to a running motion instead of a walking one For both scenarios we kept interactive framerates at all times for example when adding and removing humans from the scene or when modifying attributes such as animation colour or orientation When increasing the number of hu mans in the audience performance was dropping yet even with full capacity of the theatre around 700 places an ac ceptable rate of 27 frames per second with lighting enabled was maintained For the pedestrian crowd the performance was acceptable for up to around 500 agents simultaneously on the screen which is less than in the first scenario due B Ulicny P de Heras Ciechomski amp D Thalmann Crowdbrush Interactive Authoring of Real time Crowd Scenes to the higher complexity of models and also because of the overhead of a behavior simulation Because of the respon siveness of the crowd engine and therefore also of the inter face manipulation of the scene was immedia
21. mmon situations involving a large number of humans we created scenarios of a virtual audi ence see Figure 1 and the scenario of a pedestrian crowd see Figure 8 Virtual audience In the first scenario the tasks are given existing models of a theatre and template humans to fill the building with an audience and distribute the animations ac cording to the desire of the designer The specific scene has to be constructed according to a typical distribution of so cial classes and behaviors in antique theaters extracted from the historical literature We used four different template hu mans all with around one thousand triangles The theater model has around fourteen thousand triangles In order to facilitate positioning of the humans we created a grid of valid positions in the theatre following the distribu tion of the seats The creation brush was then restricted to operate only on this grid instead of free picking Using the grid we can position the humans without caring about colli sions for example if two humans happen to materialize very close to each other The correct position and orientation of the audience is automatic in this way the same scene will have a certain expected behavior when you interact with it much in the same way as in a paint program where pixel positions are placed in a grid Besides creation of the scene we used brushes also for interaction with the audience As soon as humans start to appear in the
22. ode the camera control is suspended and different areas on screen are selected depending on the pressed mouse button The selection areas can be in principle any arbitrary 2D shape in the current implementation the selection can be a single point a circle or a rectangle This area is then further processed by the brush according to its particular configura tion see next section The integration of the interface behavior and render The Eurographics Association 2004 ing subsystems is done using VHD a component based framework for creating real time VR systems PPM 02 Beavis Rendering Em T T OOo oe a a Operators T Selections A See a SS ee eae User Interface Ca Figure 2 Overview of the system design 3 Brushes Brushes are tools with visual representation on the screen that affect crowd members in different manners for exam ple the brush can create new individuals in the scene or it can change their appearances or behaviors We selected vi sualizations of the brushes to intuitively hint on function For example the creation brush has an icon of a human the ori entation brush has an icon of a compass the deletion brush has an icon of a crossed over human and so on The brush is processed in three stages First a selection of the affected area in the 2D screen space is performed accord ing to a triggered mouse button with subsequent picking of the entitie
23. on 2004
24. r interactive authoring followed by a detailed ac count of the concept of brushes Next we describe crowd behavior rendering and animation engines and finally we present the results where we use a prototype of the crowd brush application to create a scene of a virtual audience in a reconstruction of an ancient theatre and a scene of a pedes trian crowd in a virtual city 2 System overview Our goal is to create a system that would allow authoring of freely navigable real time 3D scenes composed of a large number of varied animated individuals in a virtual environ ment The authoring should by simple and intuitive usable by non programmers We take inspiration from the domain of image and word processing where most of the applications use WYSIWYG approaches with immediate feedback of the resulting ma nipulations In computer graphics such an approach was used for example for direct painting on models HH90 KMM 02 sketching of 3D models out of 2D drawings ZHH96 creating facial expressions PHL 98 or to paint 2D images with a virtual brush XLTPO03 A paint metaphor was used to design layouts of objects for information visual ization Tob03 and it was shown to be efficient to set values of toggle switches Bau98 Salesin and Barzel SB93 pre sented the adjustable tools metaphor where each operation is bundled with its collection of attributes to define a single tool The idea is simple the designer manipulate
25. ring approach so even pictures of real humans could be incorporated How ever Zooming on these billboards will produce aliasing arti facts due to the fact that the images on the billboards have to be small to fit in the graphics cards texture memory This makes billboarded humans a good approach for far away hu mans that do not need detailed views The Eurographics Association 2004 Culling mp Sorting mm Rasterization t Instance Information t Simulation Update Rendering Data Base Cached Keyframes Template Materials Humans Figure 6 Rendering pipeline Another approach which unifies image based and polyg onal rendering is found in WS02 They create view dependant octree representations of every keyframe of ani mation where nodes store information about whether it is a polygon or a point These representations are also able to interpolate linearly from one tree to another so that in between frames can be calculated When the viewer is at a long distance the human is rendered using point rendering when zoomed in using polygonal techniques and when in between a mix of the two It does take large amounts of data per keyframe and needs long pre processing times because of its precise nature but also gives near perfect interpola tion between detail levels without popping artifacts which otherwise occur if one uses discrete detail levels T
26. s in the 3D world space Then the operator will modify the manner of execution of the brush in the selected area Finally the brush will change the values of the instance properties for the affected individuals In case of the creation brush it will create new population members for the event brush it will send events to a behavior system and for the path brush it will add a new waypoint to a current path Values of the scalar brush parameters can be controlled by a keyboard for example sizes of waypoints for a path brush or speed of walk for an event brush are increased and decreased by pressing and keys affecting a cor responding active brush Vector brush parameters can be set by direction of the strokes so that for example a direction of movement depends on the temporal order of drawing way points Selections are defined in screen space A selection can be a single point at the location of a cursor or an area around a cursor If the selection is a single point picking in the 3D world is performed by computing the intersection of a line segment with the scene If the selection is an area picking is performed on a random sample of points from that area following a spray metaphor The size of the B Ulicny P de Heras Ciechomski amp D Thalmann Crowdbrush Interactive Authoring of Real time Crowd Scenes Figure 3 Creation brush with random operator Figure 4 Colour brush with uniform operator
27. s virtual tools working in a two dimensional screen space with a mouse and a keyboard These tools then affect the corresponding objects in a three dimensional world space see Figure 1 Different tools have different visualizations and perform dif The Eurographics Association 2004 B Ulicny P de Heras Ciechomski amp D Thalmann Crowdbrush Interactive Authoring of Real time Crowd Scenes ferent effects on the scene including creation and deletion of crowd members changing of their appearances triggering of various animations setting of higher level behavioral pa rameters setting waypoints for displacement of the crowd or sending of events to a behavior subsystem We briefly experimented with a fully three dimensional interface where tools existed in a 3D world space Never theless it appeared to be not very practical at least not when using standard input devices operating in two dimensions as a mouse or a trackball The usability of a 3D interface could be improved using some truly 3D input devices such as a spaceball a 3D mouse or magnetic sensors However it would limit the number of potential users as such devices are not common In order to create an authoring tool as outlined above the system on which it will operate should fulfil the following requirements Individuality The system must allow for each individual instance to have a set of attributes and not share them among many individuals as they c
28. state changes by sorting instances by vertex buffers and textures The most simple approach is to ren der by template then by material set and finally by instance Just before the display list is called the instance properties are applied to give the impression of variety The approximate data size of unpacked models ready for rendering is computed in the following manner Each trian gle has three vertices three normals and three texture coor dinates all together with the size of approximately one hun dred bytes For each frame the size requirement in bytes is one hundred times the number of triangles An average hu man that consists of one thousand triangles requires around one hundred kilobytes per keyframe Since we sample the animation at thirty two frames per second this amounts to approximately 3 2 megabytes per second The benefit of using display lists is that we can take ad vantage of the full OpenGL pipeline without any changes to lighting The characters look well from all camera direc tions and zooms as opposed to billboards which are usually low resolution and need pre processing for all different cam era angles The memory usage is lower than with billboards however more polygons need to be rendered The data stor age requirements on disk are very low since we store only a basic mesh and animations of the skeleton 6 Results We tested usability and responsiveness of the authoring ap plication on two very co
29. te and intuitive since changes appeared on the fly Rendering In order to measure the performance of the rendering engine we ran several tests where we varied the number of humans and the number of triangles per human The performance tests were run on a Pentium4 2 2 GHz machine with an NVIDIA Quadro4 graphics card We per formed two sets of tests depending on whether lighting was enabled or not As can be seen in Figure 7 the crowd com posed of thousand characters with 1134 triangles was ren dered at around 33 frames per second without lighting Us ing simpler characters yields a significant increase in the per formance a crowd of two thousand humans with 330 tri angles was rendered at 53 frames per second The lighting brings overhead of around twenty to eighty percent depend ing on the triangle count 7 Discussion Our authoring approach can work in conjunction with differ ent crowd rendering systems if they fulfil the requirements defined in Section 2 It would only be necessary then to define an instance properties layer abstracting implementa tion details to interface a particular crowd rendering engine with brushes CrowdBrush is also complementary to works on crowd behavior simulation where it can be used for ex ample to distribute or modify attributes and features over population of crowd agents to paint relationships among the crowd members or to send events to their behavior engines see Section 6 The m
30. y i h nidi f n ba roll Bite Giz pod OOF T hij ka W w i i j Figure 1 Crowdbrush application manual intervention intensive process operating mostly in a non real time mode A main part of the design process gen erally uses simple proxy objects that are turned into full char acters only in the final stage of production Manipulations of scenes and characters usually involve complex sequences of menu selections An interesting approach to add sets of ob jects as clouds trees flowers or buildings to the scene is used in Maya Paint Effects MAY04 where a designer can paint pseudo 3D objects in the scene using 2D brush strokes Such objects are not fully integrated into the 3D scene they are rendered in a special buffer with separate shading and are further composited into the final image as a Z buffer based postprocess Our approach is to give full creative power to design ers using metaphors of artistic tools operating on a two dimensional canvas familiar from image manipulation pro grams working in WYSIWYG What You See Is What You Get mode with a real time view of the authored scene The advantages of immediate feedback intuitive interface and fa miliarity allow to better express the artist s vision and can also lead to an increase in productivity The structure of the paper is as follows first we present the overall design of our system discussing the requirements needed fo
Download Pdf Manuals
Related Search