Real-time hand-pair segmentation and gesture recognition

This thesis presents a real-time solution to hand-pair gesture recognition using video stereo images. The solution developed here consists of three main components: segmentation, tracking and classification. The first part of the thesis addresses the segmentation of a hand-pair under different lighting and background conditions. Segmentation is achieved by using different features including color, motion, depth, symmetry and aspect ratio. These features are computed in a time efficient manner for real-time operation. Tracking of the hand-pair is then achieved by predicting the coordinates of hands after segmentation. The three coordinates of the hand-pair are extracted in order to form a gesture signature. This gesture signature is finally compared with a number of reference signatures using the dynamic time warping algorithm. The results obtained indicate a robust real-time performance of this solution in the presence of realistic lighting and background conditions.