A parallel algorithm for dynamic gesture tracking

We present a parallel algorithm for the extraction of human hand gestures from a video sequence that is able to exploit spatial and momentum coherence and color constraints using a fuzzy image integration approach. The dynamics of hand movement are critical to the understanding of gesture. In our deconstruction of hand movement streams into atomic motions we call 'strokelets', dynamic information helps in determining the roles of such movements in the gestural stream. The Vector Coherence Mapping (VCM) algorithm is used to extract the motion fields in video. These motion vectors are clustered to obtain hand motions. The parallel nature of the algorithm, and its robustness to motion blur and noise contribute to its effectiveness in gestural motion tracking. The results presented show the efficacy of VCM in the extraction of gestural motion in real video data taken under normal illumination conditions.