CS-184: Computer GraphicsLecture #19: Motion CaptureProf. James O’BrienUniversity of California, BerkeleyV2008-S-19-1.012TodayMotion Capture23Motion CaptureRecord motion from physical objectsUse motion to animate virtual objectsSimplified Pipeline:Setup and calibrate equipmentRecord performanceProcess motion dataGenerate animation34Basic PipelineFrom Rose, et al., 1998SetupRecordProcessAnimation45What types of objects?Human, whole bodyPortions of bodyFacial animationAnimalsPuppetsOther objects56Capture EquipmentPassive OpticalReflective markersIR (typically) illuminationSpecial cameras Fast, high res., filtersTriangulate for positions Images from Motion Analysis67Capture EquipmentPassive Optical AdvantagesAccurateMay use many markersNo cablesHigh frequencyDisadvantagesRequires lots of processing Expensive systemsOcclusionsMarker swap Lighting / camera limitationsCapture EquipmentPassive Optical AdvantagesAccurateMay use many markersNo cablesHigh frequencyDisadvantagesRequires lots of processingExpensive (>$100K)OcclusionsLighting/camera limitationsMarker Swap78Capture EquipmentActive Optical Similar to passive but uses LEDsBlink IDs, no marker swapNumber of markers trades off w/ frame ratePhoenix TechnologyPhase Space89Capture EquipmentMagnetic TrackersTransmitter emits fieldTrackers sense fieldTrackers report position and orientationCapture EquipmentMagnetic TrackersTransmitter emits fieldTrackers sense fieldTrackers report locationand orientationControlMay be wireless910Capture EquipmentElectromagnetic Advantages6 DOF dataNo occlusionsLess post processingCheaper than opticalDisadvantagesCablesProblems with metal objectsLow(er) frequencyLimited rangeLimited number of trackers1011Capture EquipmentElectromechanicalAnalogus1112Capture EquipmentPuppetsDigital Image Design1213Performance CaptureMany studios regard Motion Capture as evilSynonymous with low quality motionNo directive / creative controlCheapPerformance Capture is differentUse mocap device as an expressive input deviceSimilar to digital music and MIDI keyboards1314Manipulating Motion DataBasic tasksAdjustingBlendingTransitioningRetargetingBuilding graphs1415Nature of Motion DataAdjustingWhy is this task not trivial?From Witkin and Popovic, SIGGRAPH 95Witkin and Popovic, 1995Subset of motion curves from captured walking motion.1516AdjustingIK on single frames will not workAdjustingIK on single frames will not workFrom Gleicher, SIGGRAPH 98Gleicher, SIGGRAPH 981617AdjustingDefine desired motion function in partsAdjustingDefine desired function withResult after adjustmentInital sampled dataAdjustment1718AdjustingSelect adjustment function from “some nice space”Example C2 B-splinesSpread modification over reasonable period of timeUser selects support radius1819AdjustingWitkin and Popovic SIGGRAPH 95IK uses control points of the B-spline nowExample: position racket fix right foot fix left toes balance1920AdjustingWitkin and Popovic SIGGRAPH 95What if adjustment periods overlap?2021BlendingGiven two motions make a motion that combines qualities of bothAssume same DOFsAssume same parameter mappingsBlendingIf given two motions, can we blend themto find a motion 1/2 between them?Assume same DOFsAssume same parameter mappings2122BlendingConsider blending slow-walk and fast-walkBruderlin and Williams, SIGGRAPH 952223BlendingDefine timewarp functions to align features in motionDefine timewarp functionsBlendingNormalized time is w2324BlendingBlend in normalized timeBlend playback rateBlend in normalized timeBlendingBlend playback rateBlend in normalized timeBlendingBlend playback rate2425BlendingBlending may still break features in original motionsBlendingBlending may still break "features" inoriginal motionsTouchdown for RunTouchdown for WalkBlend misses ground and floats2526BlendingAdd explicit constraints to key pointsTouchdown for RunTouchdown for WalkBlendingAdd explicit constrains to key pointsEnforce with IK over time2627Blending / AdjustmentShort edits will tend to look acceptableLonger ones will often exhibit problemsOptimize to improve blends / adjustmentsAdd quality metric on adjustmentMinimize accelerations / torquesExplicit smoothness constraintsOther criteria...2728Multivariate BlendingExtend blending to multivariate interpolationBlendingExtend to multivariate interpolation"Hippiness""Speed"Weights are now barycentric coordiantes“Speed”“Happiness”2829BlendingExtend to multivariate interpolation"Hippiness""Speed"If we have other examplesplace them in the space alsoBecomes standard interpolation problem...Multivariate BlendingExtend blending to multivariate interpolation“Speed”“Happiness”Use standard scattered-data interpolation methods2930TransitionsTransition from one motion to anotherTransitioningTransition from motion A to motion BPerform blend in overlapregion3031CyclificationSpecial case of transitioningBoth motions are the sameNeed to modify beginning and end of a motion simultaneously3132Transition GraphsTransition GraphsFlipStandRunWalkSitTripDance3233Motion GraphsHand build motion graphs often used in gamesSignificant amount of work requiredLimited transitions by designMotion graphs can also be built automaticallyTransition GraphsFlipStandRunWalkSitTripDance3334Motion GraphsSimilarity metricMeasurement of how similar two frames of motion areBased on joint angles or point positionsMust include some measure of velocityIdeally independent of capture setup and skeletonCapture a “large” database of motions3435Motion GraphsCompute similarity metric between all pairs of framesMaybe expensivePreprocessing stepThere may be too many good edgesTo appear in the ACM SIGGRAPH conference proceedings2. The motion should not penetrate any objects in the environ-ment.3. The body should be at a particular position and orientation ata particular time.4. A particular joint should be at a particular position (andmaybe having a specific velocity) at a specific time.5. The motion should have a specified style (such as happy orenergetic) at a particular time.Finding paths in the motion graph that satisfy the hard con-straints and optimize soft constraints involves a graph search. Un-fortunately, for even a small collection of motions, the graph G hasa large number of edges and straightforward search of this graph iscomputationally prohibitive. The main reason is the need to enu-merate many paths. There are, in general, many perfectly satisfac-tory motions that satisfy the constraints equally well. For
View Full Document
Unlocking...