Research On Target Recognition And Tracking Methods Of Biped Robot

Biped robot system is one of the hot topics in robot field recently. The biped robot combines computer vision, motion control, sensing and data fusion, intelligent control, communication, and other related fields of technology. The vision system is an important part of biped robot and one of the most important sources that robot accesses the environmental information. Therefore, the topic has very important theoretical and practical value.The topic tried to study the robot vision system for target identification and tracking issues based on the AFU2009 biped robot research platform and the requirements of football matches. It made the robot recognize the color target real time, track and forecast target. The topic studied is as follow:1.The camera parameters of robot vision system had been calibrated. Firstly, it described the AFU2009 robot architecture system and vision system. Secondly, it analyzed the camera perspective projection model and the linear, nonlinear camera model. Thirdly, it studied the Zhang Zheng-you camera calibration method after comparison of the camera calibration algorithms. And the matlab toolbox based on the Zhang calibration method was used to calibrate the camera parameters.2.The color target recognition method of biped robot was studied. Firstly, it introduced image segmentation, histogram description of target features and other gray image processing methods, color spaces, color space conversions, region growing recognition algorithm and other color image processing methods. Secondly, the adaptive threshold update method based on shape discrimination proposed was combined with region growing algorithm to produce the improved color target recognition method in order to improve the traditional region growing algorithm. The improved method divided the image into the high and low saturation region by the S component in the HSI color space. It used region growing algorithm to identify target by H component in the high saturation region. Color threshold was updated adaptive by target shape identification. The new threshold updated old threshold in the region growing to accurately recognize the color target stably.3.The new target tracking method combining Camshift and particle filtering algorithm was proposed. Firstly, it analyzed Mean Shift, Camshift and particle filtering algorithm. In the framework of particle filter, the system state model was designed that included robot motion, camera movement and adaptive learning objectives velocity. The kernel histogram was used to describe the target eigenvalue and the degree of similarity between the candidate target area and the target area to build system observation value. Description update method of target area was used to update the initial description of target area. The best positions of particles in PF prediction were calculated by Camshift algorithm and weights of particles were updated by observation model. The state of each particle was more reasonable because of using Camshift. So it had good tracking results in the case of using fewer particles. Thus the real-time algorithm was improved greatly.