Explicit versus implicit temporal representation of temporal information in HMM based attention seeking gesture

Succesfully capturing the spatio-temporal information of gestures is the key to differentiating between attention seeking gestures and non-attention seeking gestures. Hidden Markov Models (HMM) provide a simple yet powerful mechanism for recognizing gestures with temporal and spatial variability. As simple HMMs model spatio-temporal variability as noise they are inappropriate for modeling attention seeking gestures. Two possible solutions to the problem are proposed to overcome this issue. One approach is to model temporal information in the HMM by using structural constraints in the feature space (implicit approach). A second approach is to model the temporal information in the HMM through specific structural forms in the allowable state transition nature of the HMM (explicit approach). Our experiments show that the explicit approach is superior to the implicit approach although both the explicit and implict approach better results than a simple HMM in recognizing attention seeking gestures. (Abstract shortened by UMI.)...