Research Of Target Tracking For Soccer Robot Based On Panoramic Vision

Panoramic vision camera is capable of observing 360° horizontal scene information. It overcomes the weakness of the monocular camera. In RoboC up-The World C up, the panoramic vision camera can adapt itself. Making use of the camera to obtain the characteristics of image and tracking the target rapidly, accurately and effectively, the panoramic vision camera has vital significance to the development of the follow-up control decision of the robot. According to the environment in the robot soccer matches and the characteristics of panoramic vision model, we would like to propose a video target tracking algorithm, which possesses high robustness and less restrictive conditions.Calibration of panoramic vision camera model is necessary before moving target tracking. Firstly, calibration algorithm captures the calibration pattern images in different positions by moving the camera or the calibration pattern; then extracts the coordinate of the calibration points after being preprocessed by HALCON; Finally, obtains its internal and external parameters through the expansion coefficient which are solved by the calibration points projection equation of least squares methods. This method, which doesn’t need a special device and equipment, or other priori knowledges of the scene, only needs to meet the requirements of the single viewpoint, can get a more accurate calibration. This method verifies the accuracy of the calibration by measuring the distance from the edges of the site.Taking the environment in the robot soccer matches and the characteristics of tracking the target effectively into consideration, this paper proposes an MeanShift algorithm to track the moving objects. This paper introduces the theory of algorithms and implementation steps in detail, estimates the kernel density of the moving targets by using the concept of kernel color weighted histogram, increases the robustness in the target tracking process. According to adjustment of optimum position of the step-size search moving target in the iterative process, and finally achieves the application of the algorithm in target tracking of panoramic vision soccer robot.This paper proposes some improvements for the shortcomings of traditional algorithms in real-world applications, uses Kalman filter algorithm to predict the position of the moving target, which is depend on the location of the target in the past time, speed information. As a result, the tracking accuracy of the algorithm, in the situation of fast motion and occlusion, has improved. Based on panoramic camera calibration, and the characteristics of axisymmetric relative center of the panoramic vision model, this paper divides the region of moving objectives, establishes different target models in different regions, and successfully solves the self-adapting changes of the window in the process of algorithm’s tracking. Finally this paper realizes the algorithm’s multi-target tracking in a dynamic background.The improved MeanShift algorithm applies to the target completely shelter, rapid movement, changes of shape and does well in real-time and robustness.