Curs 6

1. Curs 6 Modelare procedurală, Virtual-Humans, Modelare audio, Denavit–Hartenberg

2. Modelare procedurală • VE tind să devină din ce în ce mai complexe, pe măsură ce hardware-ul grafic 3Deste îmbunătățit și permite specificarea de detalii adiționale. • Există circumstanțe (mărimea datelor, gradul de repetițieal elementelor, structuri de obiecte particulare – spre ex. vegetația) unde este deneconceput să gândim modelarea VE folosind tehnici manuale, de scanare. • Modelarea la nivel “high-level” presupune devoltarea unor tehnici care săpună la dispoziție un nivel de abstractizarea modeluluiîn clase, care să permită specificații“high-level”. • In general, tehnicile procedurale (sau parametrice) folosesc algoritmi și proceduri pentru a coda șiabstractizadetaliile modelului, eliberândmotorul grafic de necesitatea definiriide specificații detaliate. Analysis Synthesis Parameters Hi-level model Virtual model Real model Procedures Parameters Procedures Abstraction

3. PROs • Simularea demodele complexecu mai puțini parametrii de specificat • “Data Amplification” prin controlul parametrilor • Optim pentru structuri care au un anumit grad de repetivitate • Optim pentru modelarea unei varietăți de entități similaredar nu identice, care au anumite proprietați în comun • Permit modelarea/rendarea on-demand, evitând astfel stocarea de date care nu sunt necesare (Lazy Evaluation) Procedural modelling: pros & cons CONs • Tehnicile sunt puternic legte de aplicații specifice • De multe ori nu sunt ușor de identificat, înțeles, conceput si proiectat detaliile cu privire la procedurile și setul de parametri necesari • Foarte dificil să se mențină un control suficient asupra rezultatelor

4. Paradigmele modelării procedurale Modelarea SEMI-AUTOMATĂ: Procedurile și parametrii sunt definiți, dar aceștia nu acoperă întreg procesul de generare. De aceea, în unele etape este necesară sau recomandabilă intervenția manuală a celui care modelează. Synthesis Virtual model Proc model Procedures Parameters USER USER USER

5. Modelarea procedurală

6. Modelarea procedurală a texturilor • Texturileprocedurale sunt generate algoritmic, înloc sa fie rezultatul unui proces de sampling, imagini sau desene • Exista o multitudine de abordari, folosind diferite proceduri sau parametri: • Blinn & Newell propun sinteza Fourier. • Fournier, Fussel & Carpenter propun subdiviziune fractal. • Gacalowitz a dezvoltat niste metode statistice pentru a analiza proprietățile texturi naturale a reușit să le reproducă. • Perline propune utilizarea de latice de zgomot, adică numere aleatoare sau Pseudo gradienti generate pe o grilă/matrice.

7. Latice de sunete ValueNoiseGradientNoiseValue-GradientNoise + Simple - High bandwidth + Low bandwidth - Artefacts (pattern) http://en.wikipedia.org/wiki/Lattice_Boltzmann_methods http://www.nada.kth.se/utbildning/grukth/exjobb/rapportlistor/2004/rapporter04/eriksson_erik_04161.pdf

8. Perlin Noise • Es. 2D PerlinTurbulence: Value = 0 for (f = MINFREQ; f<MAXFREQ; f*=2) value += 1/f * noise(P*f)

9. Texturi procedurale • Pros: • Bandwidth and memorysaving (no storageneed) • No tiling • High detailindependentfrom zoom • Procedures: • IBR (Image Based-Rendering): textures are generated starting from samples • Texturesynthesis: texture are generatedstartingfrom material properties • IBR pro: no needtoidentifydifferent pro- ceduresfordifferentmaterials, storeonlysmallssamples

10. IBR →

11. IBR + algoritmi genetici http://www.youtube.com/watch?v=Fp9kzoAxsA4

12. Procedural Texturing - Animation • Real-time animatedtextures • Allowsproducingcomplesanimations on images and not on polygons (es. facialanimation, fire or liquids, clouds etc.)

13. Procedural synthesis of geometry • L-Systems: FormalgrammarproposedbyLindenmayer (1968) and adaptedtographicsbyAlvy Ray Smith (1984) • An L-System isbasedupon: • An alphabet (e.g. “F”, “+”, “-”) • A set of production rules (e.g. F → F+F--F+F) • Productions are applied in parallel, i.e. the biggestnumberofsymbolsissubstituted at eachinstance (output isindependentfromrulesapplicationsorder)

14. Procedural synthesis of geometry • Symbols can havegeometricalmeanings • Es. Logo: • “F” draws a segment • “+” rotatesθdegreescounterclockwise • “-” rotatesθdegreesclockwise http://en.wikipedia.org/wiki/Fractal

15. Procedural synthesis of geometry • L-systems are more flexibleintroducingpush and popoperations (symbols “[“ amd “]”) • Thisallowtorealizeverycomplexstructures: • F → F [+F] F [-F] F

16. Procedural synthesis of geometry • Passingto 3D: • “+” e “–” foryaw • “^” e “&” forpitch • “\” e “/” forroll • In 3D segmentsmayberepresentedbycylinders or conefrustums

17. Paradigmsofproceduralsynthesis • The typicalmodeling data flow is: • The userspecifies a conceptualmodelto the modeler • The modelerconverts the conceptualmodel in an intermediate representation, suitableforbeingprocessed and rendered • The renderertakes the representation and synthesizeanimage • Proceduralsynthesisrelies on twodifferentparadigmstospecify the intermediate representation: • Data Amplification • LazyEvaluation MODELER UTENTE RENDERER Concept Geometry

18. Data amplification • A high geometricaldetailisspecifiedstartingfromfew input information • L-Systems are a typicalexampleof data amplification (es. poplartree 16 Kb (concept) -> 6.7 MB (polygons) • Data explosion: high memorystoragerequirements MODELARE(Amplificare) USER RENDERER Articulare Geometria

19. Lazy evaluation • Thisparadigmavoids the intermediate representation, generating on the fly the synthesisonlywhenneededforrendering • Low storagerequirement, high real-time demands MODELARE(server) USER Geometria RENDERER (client) Concept Coordonate

20. ParametricL-Systems • ParametricL-systems can change the behaviourdepending on the passedparameters. Thisallowsto alter transformations and toallowrecursion • Es. inductiveinstancing define grass(0) < iarbă > definegrass(n) <grass (n-1) grass (n-1) translate 2^n * (0.1, 0, 0) grass (n-1) translate 2^n * (0, 0, 0.1) grass (n-1) translate 2^n * (0.1, 0, 0.1) >

21. Procedural Geometric Instancing • Global coordinates: An instancemayvaryitsgeometrydepending on its world position/orientation Es. Tropism http://en.wikipedia.org/wiki/Tropism es. top → down: gravity bottom→ up: searchingsunlight side: wind, etc.).

22. Virtual Humans

23. Avatar • “Icon or interactiverepresentationof a user in a sharedenvironment” • “avatar” comesfrom Hindu, describinggodVishnuembodiement • in text-based VR (suchasMUDs) an avatar is a textdescriptionprovidedtootheruserslooking at the user • In graphics VR, an avatar is a 3D model of a virtualhuman (or character) directlydrivenby a humanuser

24. Agents • An agentisanautonomous VH whoseactions are notdrivenby a humanbeingbutratherby a computer • VH actions can beguidedby: • anauthonomoussensory system • behaviouralrules • predefinedscripts

25. VirtualHumans: requirements • Graphicssimulation: • Rigidbodies • Deformablebodies • Physicalsimulation: • Physicallybasedmodeling • Inverse and directkinematics • Behavioural simulation

26. GraphicalrepresentationofVHs • Layeredmodeling: • Simplest case: rigidbodieshierarchy (no deformations) • Usually at leasttwolayers: skeleton and skin • More layers: anatomybasedapproach • VH Animation • Skeleton motion • Consequent skin blending

27. Reprezentare grafica a VHs • Skeleton + skinning

28. Skinning • Simpledeformationalgorithmallowing a skeletontocontinuously animate anassociatedskinmesh • Eachskinvertexisinfluencedbyone or more bonesof the skeletondepending on a system ofweights. • Toconnect the skinto the bones (rigging) envolopesofpredefinedshape and size can beused, withaninternal and anexternalboundary • Vertices inside the internalboundary are givenweight = 1 • Verticesoutside the externalboundary are givenweight = 0 • Verticesbetween the twoboundaries aregiven a weightbetween 0 and 1 • A vertex assume the position relatedto theenvelopeitisincludedinto • If multiple envelopes include a vertex, itwill assume an intermediate position givenby the weightedaverageof the positions vblend= v1w1 + … + vn-1 wn-1 + vn (1-Σiwi) Blending formula

29. Rigid bodies vs. skinning • Rigidbodieshierarchy: • P: Simple, lightweight • S: Imprecise, doesnot simulate deformations It can be considered a borderline case of blending, where each vertex is associated to one bone only with weight = 1

30. Skinning – Bind Pose • When the skinislinkedto the skeleton, the “Bind pose” issaved, whichis the status, in world-space, of the transformationmatricesof the bones and of the skin, whichwillbeused in the blending stage. • Duringblendingeachskinvertex: • Istransformedfrom the skinlocal-spaceintoworld-space • The fromworld-spaceto the localspaceofeachboneitisconnected • Itundergoesall the animationsofeachboneitisconnected, and foreach a correspondingnew positioniscomputed • Alle the newpositions are transformedintoworld-space and thenweighted

31. Skinning – Vertex shader • Whendealingwithcomplexskins and skeletons, blending can be CPU intensive • The sameoperation can beperformed on GPU byprogrammingan opportune vertex shader • Important: in orderto produce a correctlighting, normalsmustbeblendedtoo!

32. Animationtechniques

33. Forwardkinematics • A human body can be sketched as a connected set of: • links (arm, forearm, etc.) • joints (connecting links: elbow, shoulder etc.) • In order to animate a skeleton, it is needed toalter the joint angles: • In a 2-layers representation, only the bones are animated composing, each frame, a particularposture, corresponding to a particular configurationof the joint angles array • The skin is then blended accordingly • Needed info: • Anatomy based joint angles limits • Masses of links (only for PBM) http://www.youtube.com/watch?v=VjsuBT4Npvk&NR=1&feature=endscreen http://www.youtube.com/watch?feature=endscreen&NR=1&v=3ZcYSKVDlOc http://en.wikipedia.org/wiki/Denavit%E2%80%93Hartenberg_parameters

34. FK: MotionCapture • Joint angles can be computed (Motion Synthesis) or sampled using sensors (Motion Capture) • Motion capture pros: • Realistic animation • Little work for modelers(only fine tuning and junctions) • Captures animation detailsthat are uncatchable from theeye or difficult to synthesize • Cons: • Requires expensive devices • Data are samples, which aremore difficult to process

35. Inverse kinematics • The kinematicsof a connectedstructureis the processofcomputing the position in the spaceof the structureend-effector, given the jointangles • Inverse kinematics (IK) is the oppositeprocess: given theend-effector position (goal or target), retrieves theconfigurationof the related joint angles. Itis a processwidelyused in robotics. • Pros: • Allowstocalculatedjointsstartingonlyfrom someposition information (typicallyhands, feet, head), reducing the numberofneededsensors • Cons: • In some “singularity” point more thanonesolutionispossible. Althoughanatomicallimits can help, itisnotalwayspossibletofind the correctone. http://demonstrations.wolfram.com/ForwardKinematics/

36. Shape Interpolation • Two or more 3D reference meshes representing human body postures are morphed • Generally the needed steps are: • 1 – Meshes are morphed • 2 - Texture coordinates are morphed • 3 - Texture maps are morphed • If meshes are related to the same basic shape, steps 1 and 2 arebasically linear interpolations. • Step 3 is needed only for some visualeffects

37. Keyframe Interpolation • Sometimes in literature is synonim of Shape Interpolation • Our definition refers to Skeletal Animation • Reference values are not diretly meshes, rather skeleton postures related to particular keyframes • Interpolations takes place on these values, determining new skeleton posture • The skin is blended frame by frame based on the new interpolated skeleton postures

38. Physicallybasedmodeling • Animations based on kinematics do not keep into account physical effects like gravity or inertia • Rather than setting up a kinematics problem, we can setup a dynamics problem, considering also masses and forces • Inverse dynamics is also possible (for each joint, computes forces and torques generating a desired movement) • Modeling dynamics is DESIRABLE to correctly manage the interaction between the VH and the VE (collision detection and management etc.)

39. Behaviouralmodeling • VHs must be able to access information about the VE, either directly (unrealistic) or by means of a system of virtual sensors • These informatino will drive his actions: (behavioural modeling) • locomotion driven by sight • object manipulations • feedback to acoustic stimula etc. • Behaviours can be scripted or procedurally evaluated • Other behavioural issues: • Interaction among VHs • Interaction between VHs and real humans

40. Standards

41. H-Anim • HumanoidAnimation: • Target: creationof a libraryofinterchangeablehumanoids and authoringtoolsto create newhumanoids and animations • Support keyframe, IK, FK, etc. • http://www.h-anim.org/Specifications/H-Anim1.1 • Features: • VRML 97 compliant • Flexibility, no assumption on the applicationtype • Simple (dealsonlywithVHs and no otherarticulated figure)

42. H-Anim – File format example ... DEF hanim_l_shoulder Joint {  name "l_shoulder"   center 0.167 1.36 -0.0518   children    [     DEF hanim_l_elbow Joint {  name "l_elbow"       center 0.196 1.07 -0.0518       children        [         DEF hanim_l_wrist Joint {  name "l_wrist"           center 0.213 0.811 -0.0338           children    [             DEF hanim_l_hand Segment {  name "l_hand“       ...             }           ]         }        DEF hanim_l_forearm Segment {  name l_forearm"            ...         }       ]     }     DEF hanim_l_upperarm Segment {  name "l_upperarm"       . ..     }   ] } ...

43. H-Anim - Hierarchy

44. MPEG-4: body animation • FBA: Facial Body Animation in Mpeg-4 • An Mpeg-4 body is a collectionofnodes • Root, BodyNode, contains 3 nodes: • BAP (Body AnimationParameter): 296 parametersdescribingskeletonproperties • Rendered Body: holds DEFAULT skin information (shape + textures) • If a specific body mustberendered, the BDP (Body DefinitionParameters) isadded, replacing the defaultdelRendered Body. Skinningparametersmaybespecified. • The H-Animgroup, coordinatedwith theFBA groupof Mpeg-4 hasstandardizedVH specifications, so asto produce coherentresults in bothennvironments.

45. Virtual Crowds

46. Virtual Crowds • VHs are neededtopopulateVEs • Some VEs (suchascities) need a high numberofVHs • Crowdssimulation, becauseofitsintrinsiccomplexity, cannotbemanagedas sum ofVHs • Application: entertainment, studyofcrowdflows (panic, disaster etc.) • Requirements: • Graphicsmodeling • Behaviouralmodeling

47. Virtual Crowds

48. Virtual Crowds example • Real-time oriented • Allowsto render ~ 100K differentVHs • VHsrenderedasprecomputedimpostors • Several VH types, possibilityofmodulatingcolors andtextures • Real-time shadowingusingshadowmaps

49. Virtual Crowds • Behaviouralalgorithms: • Collisionavoidance • Heightcheck • Interest attractor • Exit search, visibility, flow inertia etc.

50. AcousticalEnvironment • When a sound isgenerated, itispropagatedaswaves in a medium. The propertiesof the medium, and of the surroundingenvironment, influencehow the sound isperceived • A complete acousticalfieldiscomposedof: • Sound sources • Listener • Environment