Hand tracking by combining enhanced incremental learning and background model

Hand tracking is a difficult problem because of its highly articulated characteristic and complexion-like disturbance. This paper proposed an enhanced incremental subspace learning (EISL) algorithm for color image. In this method, the HSV color space is used in consideration of its individuality, clustering and compatibility to human color perception, incremental subspace learning used for tracking is based on high dimensional vectors reshaped from the three color channels. Considering tracking failure, a dynamic background model is established and applied to deal with the problem. Experiment results show that our tracking algorithm is robust for viewpoint change, distortion, drastic illumination change and partial occlusion, and the method for tracking failure judgment is effective.