An Improved Hand Tracking Algorithm for Chinese Sign Language Recognition

In the process of sign language recognition, whether the tracking algorithm can accurately track the hand directly determines the final recognition result. The KCF(kernelized correlation filter) tracking algorithm is generally used at present. It uses the kernelized correlation filter to detect the position of the tracked target. However, it can only track single-scale targets and does not handle the occlusion problem well. In sign language, the size of the hand being tracked changes constantly. Deformation of the hand can also result in partial or even complete occlusion. Based on the traditional KCF tracking algorithm, a method to establish an independent scale filter for multi-scale tracking is proposed. Kalman filtering algorithm and adaptive model updating strategy are introduced to solve the occlusion problem. Experiments show that the improved algorithm can detect the position and scale of the hand effectively and track the hand well under occlusion.