Hand Gesture Tracking Based On Color Image And Depth Image

Using computers to track and identify the hand gesture is an important method of nature Human Computer Interaction.Currently,great progress has achieved in the research of hand gesture recognition,but is still a long way to the practical application.Many gesture recognition method based on color images using a variety of algorithms,however,often be identified for some particular gestures,and lack robustness in complex backgrounds.Depth images provide a different kind of data for gesture recognition algorithm,which are usually better in eliminating the interference of the environment,because it is not subject to the impact of light and the farther environmental background can be easily removed.Therefore,this paper presents a color image and a depth image-based gesture tracking method,which can identify any hand gestures rather than just a specific gesture,and can be run in real time online.Using random forest algorithm for skin color classification in the color image;Using the same random forest algorithms for segmentation of the hand in the depth image.A simplified 26 degrees of freedom 3D hand model is adopted and an objective function which calculates the difference between the hypothesis hand model gesture and real hand gesture is presented.Particle swarm optimization algorithm is utilized to solve the objective function,and the GPU is adopted to accelerate the particle swarm optimization.