Real-time hand-pair gesture recognition using a stereo webcam

This paper discusses a real-time hand-pair gesture recognition solution using a stereo webcam. The developed solution consists of three main components: segmentation, tracking, and classification. Segmentation is achieved by combining color, motion, depth, symmetry and aspect ratio features. All these features are found in a computationally efficient manner to allow real-time operation. Tracking is done to relieve the computational burden of segmentation for every frame. The three coordinates of the detected hands are then used to form a hand-pair gesture signature which is compared to a number of reference gesture signatures using dynamic time warping. The results obtained indicate a robust real-time performance of this solution in the presence of realistic lighting and background conditions.