Real-time hand-arm motion analysis using a single video camera

We describe a video-based analysis system for acquisition and classification of hand-arm motion concerning German sign language. These motions are recorded with a single video camera by use of a modular framegrabber system. Data acquisition as well as motion classification are performed in real-time. A colour coded glove and coloured markers at the elbow and shoulder are used. These markers are segmented from the recorded input images as a first step of image processing. Thereafter features of these coloured areas are calculated which are used for determining the 20 positions for each frame and hence the positions of hand and arm. The missing third dimension is derived from a geometric model of the human hand-arm system. The sequence of the position data is converted into a certain representation of motion. Motion is derived from rule-based classification of the performed gesture, which yields a recognition rate of 95%