eg Sign language → Highly structured More on Communicative Gestures.pdfVIP

Review of Gesture Recognition Techniques By François Rioux, Feb. 23, 2004 Presentation Outline Overview of the problem Human gesture representation Features selection Recognition techniques HMM and « improved » HMMs Other techniques Gesture Recognition = Complex Task Motion modeling Motion analysis Pattern recognition Machine learning Psycholinguistic studies … Human Gesture Representation Psycholinguistics research by Stokoe: Hand shape Position Orientation Movement Human Gesture Representation(cont) Application scenarios of gestures Conversational Controlling eg: vision-based interfaces Manipulation eg: Interact with virtual objects Communication eg: Sign language →Highly structured More on Communicative Gestures Highly structured 3 phases Preparation Stroke (dynamic part) Retraction The dynamic part contains the information Features Extraction Good feature extraction is CRUCIAL Static hand posture Fingertips Finger direction Hand contour … Features Extraction (cont) Dynamic of gestures Spatial features eg: 2D location of hands. More general: 3D Temporal features Need of statistical method independent of time Choosing only most relevant features MEF: Most Expressive Features Karhunen-Loeve projection MDF: Most Discriminative Features Temporal Gesture Modeling and Recognition Similar to speech recognition HMMs can be used for low level recognition However more complicated than speech… Gesture semantics for high level recognition (Hidden Markov Model Overview) Modeling the Dynamics Low-level dynamics of human motion Useful for human motion recognition Quantitative representation of simple motion Not sufficient for complex moti