Vision-based hand gesture spotting and recognition

In this paper, we propose a dynamic gesture spotting and recognition algorithm using our stereovision system. The 3D trajectories of hand gestures are first reconstructed by a stereovision-based motion capture platform. Hand gestures can then be segmented from the trajectory in real time by using proposed gesture spotting algorithm. Discrete cosine transforms coefficients, complex index and gesture entropy features are extracted to represent the gestures. With these features, one-class SVM is adopted for gesture classification. The experimental results demonstrate the feasibility of proposed spotting and recognition algorithm.