Research On Target Detection And Tracking Of Robot Based On Binocular Stereovision

Vision technology of mobile robot is one of the hot topics in robot field recently. Just like human beings, for the mobile robot, vision is considered as one of the most important ways to get the information of surroundings. With the fast development of computer and electronic technology, the binocular stereovision could meet the requirement of real-time applications and will be widely applied in the navigation of robot, the production automation, etc. Therefore this subject has the important theory significance and the practical value.This article is studying the basic algorithms of binocular stereovision based on Marr's vision theory. The AS-R mobile robot's binocular vision system is taken as a platform based on which we do research about the tracking and location of the object. The contents studied are as follow:Firstly, the essay introduced the principle and structure of binocular vision system then we described the structure of the AS-R robot system, and binocular vision system was extended. We did some research on the existing camera calibration methods. Finally, the Zhang calibration method was used to get the parameters of binocular stereovision.Secondly, the color object recognition and tracking algorithm based on Kalman Filter was proposed. The algorithm implemented rapid image threshold segmentation based on color information in the HSI color space and obtained the pixels of the object. The "four key points" method was used to estimate the center and radius of football. An adaptive threshold update was proposed to improve the robustness of the algorithm. The single pixel football edge was extracted by Canny operator. In order to solve the target tracking problem, the possible position of the moving object in the next frame was predicted by Kalman filter, with the motion of robot as the external control of Kalman filter. So the target was identified and football edge detection only in the predicted area advancing the real-time of algorithm.At last, eight edge points were extracted almost evenly in the edge points which had been detected. After that, we found eight match points in another image. So far sixteen match point pairs which had been chosen by a two-way matching of two images were extracted to meet the requirement of three-dimensional match. Parallel optical axis binocular vision system's principle of 3D object location was analyzed. The 3D coordinates of sixteen points on the football edge could be obtained easily by binocular vision system of parallel optical axis and inverse transform of camera imaging in this subject. The average of these coordinates was the 3D coordinate of football core, and then the football target's position was known.