Robust Object Tracking Technology Based On Feature

Target tracking in terminal guidance is the key technique in a precise guidance weapon system that ensures the attacking accuracy of precise guidance weapons.Widely used in modern wars,precise guidance weapons can even change the outcome of a war.Therefore,this paper focuses on two problems in target tracking of terminal guidance:In the phase of terminal guidance,a method is needed to acquire the accurate target location in real-time images as real-time images and baseline images are quite different.A target locating algorithm based on visual angle compensation was proposed to simulate baseline image in different angles through perspective transformation and locate the target by image matching with the aid of local invariant feature description.In the phase of target tracking,as the size of target increases during the missile approaching the target in the phase of terminal guidance,tracking algorithm is required to accommodate the change of size of target.Thus a compatibility matching algorithm was put forward.In this algorithm,the BRISK feature is selected to track based on keypoint matching,which is invariant to target rotation and scale variation.Two complementary strategies are adopted to find the keypoints.One of them is used for point matching,and the other one is used to eliminate unmatched keypoints based on a forward and backward optical flow method.This method can effectively increase the matched keypoints pairs.The target angle and scale variation are estimated by the consistent matching of the keypoints.Target position and size are estimated by a voting mechanism.As there are error keypoints pairs,a clustering method is proposed to analysis the voting results,and a distance threshold is used to eliminate the anomaly voting keypoints and its corresponding voting results,which is simple and efficient.Experimental results show that our algorithm can accurately estimate the target size and angle variation in the tracking process,and achieve the more stable tracking results.