Research On Gesture Recognition And Tracking Based On Binocular Vision

One of the hottest topics in the field of human-computer interaction(HCI) is how to operate computers by a more effective and natural way in order to reach better user experience. Because of human hands contain a large number of interactive information with human cognitive habits, gesture-based HCI system is more acceptable for human users than other HCI systems. Besides, for the continuous movement of hands, the gestures also contain large number 3D space position information. So it is necessary to combine gestures space position information and the shape information as a whole. In addition, the vision-based gesture recognition and tracking technology with low hardware prices, non-contact characteristics and market prospects.Although many scholars have studied the gesture recognition and tracking technology based on binocular vision, but there are still many problems, such as hand detecting and tracking have low accuracy and professional binocular vision equipments are expensive. Through literature research, this paper researched some important topics on binocular vision gesture recognition and tracking technology. The main works are as follows:Firstly, analyzed the mixtures of multiscale deformable part model and created the directly hand shape detector and the indirectly hand context detector. Also we proposed the bilateral skin color extraction detector based on the adaboost face recognition result. Then designed a new kind of hand detect method which named multi-detector hand recognition method by the above three kinds detector.Secondly, to solve tracking candidate features single and easy lost high speed moving targets, this paper improved the color histogram-based Mean Shift tracking algorithm by combining the LBP features and Kalman filter to match the scene of gesture recognition, which has a good performance on fast move hand tracking.Thirdly, combined with the principle of binocular stereo vision and camera calibration related outcomes, we designed a binocular stereo matching ranging systems based on FAST features. We finished the camera calibration works and tested the system. The test result shows that the relative error of this system is less than 6% when the system and targets distance within 800 m.Considering the gesture recognition and tracking is a practical technology, so we combined the above findings, in this paper, designed an experimental binocular vision based gesture recognition and tracking final system. Upon the examination of this system, it is able to extract the gesture binocular camera system before the 3D shape information and location information, and has a certain practical significance.