Dynamic Hand Gesture Recognition Using Hidden Markov Models

Hand gesture has become a powerful means for human-computer interaction. Traditional gesture recognition just consider hand trajectory. For some specific applications, such as virtual reality, more natural gestures are needed, which are complex and contain movement in 3-D space. In this paper, we introduce an HMM-based method to recognize complex single hand gestures. Gesture images are gained by a common web camera. Skin color is used to segment hand area from the image to form a hand image sequence. Then we put forward a state based spotting algorithm to split continuous gestures. After that, feature extraction is executed on each gesture. Features used in the system contain hand position, velocity, size, and shape. We raise a data aligning algorithm to align feature vector sequences for training. Then an HMM is trained alone for each gesture. The recognition results demonstrate that our methods are effective and accurate.