Research On Gesture Recognition Based On Kinect And Human-robot-interaction Systems

Due to the visual interaction is worked in the real life scenes which is changeable andunpredictable and the scene will be influenced by a lot of interference from external factors, suchas light, temperature, so the current vision-based human-robot interaction is limited to a certainextent. Aiming at this problem, a gesture recognition method based on depth sensor is presentedin this paper which can overcome the shortage that the traditional hand gesture recognition iseasily affected by illumination changes. A gesture recognition-based human-robot interactionframework oriented to mobile manipulator platform is proposed and the intelligent level of themobile manipulator platform is enhanced.The present situation of research on visual gesture recognition and human-robot interactionare analyzed in this paper and aiming at the existing problems of gesture recognition at present,the solution of gesture recognition and human robot interaction based on somatosensory sensorKinect is proposed.Firstly, the basic concept of gesture is expounded and the common methods of gesturemodeling, static gesture recognition and dynamic hand gestures recognition are summarized andanalyzed. The skin color modeling method based on Single Gaussian Model and GaussianMixture Model are discussed successively and the color detection experiment is carried out.Secondly, the principle of obtaining depth image with somatosensory sensor Kinect isanalyzed specially and the method of gesture pose and articulation points based on depthinformation are studied. Then3D modeling of gesture combined with depth information and skincolor model is presented. On this basis, gesture feature extraction based on Haar-Like feature isrealized which establishes a good foundation for the subsequent gesture recognition.Thirdly, the principle of Kalman filter and Camshift algorithm are described successivelyand an algorithm combined with Camshift tracking algorithm and Kalman filter algorithm isproposed and the accuracy and instantaneity of dynamic gesture tracking are improved.Finally, the static gesture set and dynamic gesture recognition experiments are carried outand according to the recognition results, the performance such as recognition rate, stability andinstantaneity are discussed. Then the common model of human-robot interaction is analyzed anda gesture recognition-based human-robot interaction framework oriented to mobile manipulatorplatform is proposed and the control gesture set and the system function are designed. On thisbasis a navigation experiment based on human-robot interaction is carried out and theeffectiveness and accuracy of system are verified.