Tracking And Navigation Of Mobile Target Based On Image

With the continuous development of the computer vision and digital image processing technology, mobile target tracking technology based on vision has been widely used in the mobile robot, video monitoring, aerospace and other fields. But the existing tracking algorithms are only applicable to specific experimental scene because of the complexity of the scene and variability, therefore, study of target tracking technology still has a high practical significance. This paper made a thorough study of algorithms for the detection and tracking of moving targets, and the improved algorithm proposed had applied to multi-robot cooperative hunting experimental system, provides real-time location and navigation information for the robot. Verified by the experiment that this algorithm has good real-time performance and accuracy. The main work is as follows:First of all, commonly used object detection algorithm including frame difference method, background modeling algorithm and Grabcut image segmentation algorithm is studied, and according to the experimental results that Codebook algorithms extracted moving targets exists shadow interference problem, Codebook based on threshold segmentation algorithm is proposed, improving the segmentation result and real-time performance of the algorithm.Secondly, the Camshift tracking, Kalman filtering and compression tracking algorithm based on compression perception were studied in this paper, and aiming at the shortcomings of the Camshift tracking, presented the algorithm combining with Kalman filtering and Camshift, the algorithm improves the ability of anti-interference of the same color. At the same time, the research on median flow tracking algorithm based on image pyramid was done, and the algorithm increased forward-backward trajectories verification based on the image pyramid optical flow, not only has good real-time performance, but also has a strong light anti-interference ability and so on.Finally, the paper introduced in detail about our experimental platform of the overall system architecture, including software and hardware platform and the overall work process. Also, because the experimental system to the mobile robot navigation information requires precise target location, this paper used the direct linear transformation to obtain internal and external camera calibration parameters, then converted the coordinates of the target image to the actual scene coordinates. There still exist images flash and camera overlap processing problem in the actual system application, this paper presented the solution which is suitable for this experiment platform after a series of studies. The experiment verified that, the Codebook background modeling algorithm and the median flow tracking algorithm based on image pyramid selected by this article, has a good real-time and accuracy and meet the needs of the actual system.