Upload
sharne
View
74
Download
0
Embed Size (px)
DESCRIPTION
Finger Detection for Multi-Touch Tabletop Display System 多重觸控桌面顯示系統之手指偵測. Su-ting, Chuang 2010/8/2. Outline. Introduction Related Work System and Method Experiment Conclusion & Future Work. Outline. Introduction Related Work System and Method Experiments Conclusion & Future Work. - PowerPoint PPT Presentation
Citation preview
Finger Detection for Multi-Touch Tabletop Display System
Finger Detection for Multi-Touch Tabletop Display System
Su-ting, Chuang2010/8/21OutlineIntroductionRelated WorkSystem and MethodExperimentConclusion & Future Work22OutlineIntroductionRelated WorkSystem and MethodExperimentsConclusion & Future Work33IntroductionNon-uniform lighting problemVarious finger touch response among different positionLow computation efficiencyNo such tool that helps users determine parameters automatically
44OutlineIntroductionRelated WorkSystem and MethodExperimentsConclusion & Future Work55Related WorkFTIR (Frustrated Total Internal Reflection)
J. Y. Han, Low-cost multi-touch sensing through frustrated total internal reflection," in Proceedings of the 18th annual ACM symposium on User interface software and technology (UIST '05). New York, NY, USA: ACM Press, 2005, pp. 115-118.
66Related WorkDI (Diffused Illumination)J. Rekimoto and N. Matsushita, Perceptual surfaces: Towards a human and object sensitive interactive display," Workshop on Perceptural User Interfaces (PUI'97), 1997.
7An IR camera with IR illuminators to observe hands
7Related WorkTouchLibA multi-touch development kit
Finger detection processing flow chart8
BackgroundSubtractionSimpleHighpass
ScaleThresholdFinger Analysis background (edge)Scale thresholdthreshold8Related WorkDirectShowFilter-based framework GShowGPU-accelerated frameworkCombination of DirectX and DirectShow
9Dshow: [media-streaming] Provide high-quality capture and playback of multimedia streams can perform high-quality video and audio playback or captureIt supports capture from digital and analog devices based on the Windows Driver Model (WDM) or Video for Windows. It automatically detects and uses video and audio acceleration hardware when available, but also supports systems without acceleration hardware.
Gshow: combination of directX and directShow
9OutlineIntroductionRelated WorkSystem and MethodExperimentsConclusion & Future Work1010Hardware Configuration
(2) IR Camera(3) IR Illuminator(1) Peripheral Projector1111Hardware ConfigurationOrder of diffuser layer and touch-glass layer12
Diffuser layerIR illuminatorIR cameraspotIR illuminatorIR cameraTouch-glass layerIR cameraspotIR camera
2143The surface of this kind of tabletop system is usually composed of a diffuser layer and a touch-glass layer.12Hardware ConfigurationProblem:IR rays reflected by the touch-glass will result in hot spot regions in camera viewsSolution:Use other cameras to recover the regions which are sheltered by IR spots
13
camera13Software ArchitectureDetection systemImage StitchingFinger DetectionFinger Tracking
Parameter determination14ImageStichingFingerDetectionFinger TrackingImage StitchingCombine multi-camera view into a virtual camera viewFinger DetectionRecognize touched fingertipFinger TrackingSmooth the trajectory of finger and fix lost results14Software Architecture15
ImageStichingFingerDetectionFinger Tracking15Image StitchingGoalCombine multi-camera view into a virtual camera view
16
Image StitchingAdvantagesRemove IR spot effectUnify finger size among different position of tableReduce matching problemBe compatible with existent finger detection system
17
Image Stitching18ImageBlendingIR Camera(L)IR camera(R)UndistortionUndistortionHomoWarpHomoWarpImage StitchingHomoWarp19
521625431436
2519Image StitchingImage Blending
20
20Finger DetectionTouchLib
Our method21NormalizationDifference of GaussianBackground SubtractionBinaryFinger AnalysisSimple HighpassScaleBackground SubtractionBinaryFinger Analysis
Finger DetectionNormalizationMethodModel distribution of IR illuminationUse specific material to simulate foregroundConstruct normalization map
Normalize foreground image
ResultBefore normalization: mean = 75, standard variation = 30After normalization: mean = 255, standard variation = 322
Construct normalization mapCalculate each pixels dynamic range Stretch dynamic range to 0-255Normalize foreground imageMultiply normalization ma
22Finger DetectionDifference of Gaussian (DoG)Modified from simple highpass in TouchLib
23
Simple highpass -blur, noise, touchLibmedian filter , midian filter time complexity
sigma = (n/2 - 1)*0.3 + 0.8 , n=kernel size23Fingertip TrackingGoalSmooth the trajectory of finger Fix lost resultsMethodKalman filterSmooth the path
Predict the new state and its uncertaintyCorrect the tracker with its new measurementAssume white noise and uniform velocityOriginalAfter Kalman filter2424Parameter DeterminationRequirements of ideal finger detection systemSensitive missNoise-free false alarm GoalFind an applicable set of parameters for finger detection system fulfilling the requirements25Parameter Determination26Parameters DeterminatorParameter CombinationDetection ResultApplicable set of ParametersTestSetTouch DataGround Truth(Trace)Detection SystemParameters Determinator : generate parameter combinations & evaluate parameter combinations by calculating miss and false alarm
26Parameter DeterminationEvaluation of parametersData CollectionDepict traceMeasurementMinimize # of miss and false alarm27
requirmentsmiss & false alarm27Parameter DeterminationIdeal finger detection Only one fingertip landing on traceContinuity among frames28
Good frameOnly one finger landing on traceContinuity
28OutlineIntroductionRelated WorkSystem and MethodExperimentsConclusion & Future Work2929ExperimentsPerformance evaluation30
30ExperimentsParameter determinationDecide parameters in our systemAdopt sampling-based parameter search technique
31NormalizationDifference of GaussianBackground SubtractionBinaryFinger AnalysisSubtract valueSmoothkernelThresholdFingerSize31ExperimentsParameter determinationExhaustive searchParameter combination5 (step) *5 (step) *5 (step) *5 (step) = 625Applicable parameter num16/625 = 2.56%
32Subtract valueSmooth kernelThresholdFinger sizeLow bound051010Step55510High bound2025305032ExperimentsParameter determinationParticle filtering33SamplingMeasure
Initialize particlesParticle generationWeight Computation12M12MNormalize WeightsMore iterationOver limited iterationResamplingoutputExitNoYesYesNoThe key idea is to represent the required posterior function by a set of random samples with associated weights and to compute estimates based on these samples and weights.
33OutlineIntroductionRelated WorkSystem and MethodExperimentsConclusion & Future Work34343535