Both hands pose recognition control system based on skeleton tracking

This paper presents a method for two-hand pose recognition based on skeleton information, aiming at the problem of low recognition rate and poor robustness in the field of human-computer interaction by single hand. This method consists of two steps: two-hand positional information extraction and gesture recognition. In the first step, we utilize the Kinect depth image to acquire the position of both hands. The second step is the highlight of the proposed method, it locates the palms by hand nodes, extracts the right hand movement information which is trained by a Hidden Markov Model. This method has been verified by a experimental car control system and demonstrated good robustness in complex background environment.