Real-time multi-cue finger tracking for human computer interaction

In this paper we propose an algorithm for finger detection and tracking in video sequences captured with a common webcam. The proposed method uses a multi cue approach to achieve finger detection and significantly reduces the amount of data to process by concentrating the results of preprocessing steps in a reduced number of line strip features. Finger candidates are identified based on quasi contiguous clusters of line strips through multilayer processing. Position, size and angle parameters are extracted for the valid finger candidate. The finger parameters obtained in the current frame are used to predict the finger parameters in the next frame and can also be used within a user interface either for pointing or for dynamic gesture recognition. Tracking performance is analyzed in challenging situations including fast motion and occlusion from similar objects. The proposed algorithm has a computationally effective architecture and is suitable for realtime implementation on systems with less complex processors.