Target Tracking System For UAV Based On Stereo Vision

Target tracking is an important research and application direction of UAV.In this paper,a long-time target tracking algorithm based on binocular vision was designed for the problems of complex target changing,changing tracking scene and unstable sensor signals in the target tracking mission.Design experimental verification this algorithm can be used for pedestrians,vehicle tracking,guided landing,target guidance and other scenarios.The main work and contribution of this paper are as follows(1)In this paper,the theoretical knowledge of L-K optical flow method and Sift feature point matching are reviewed.The principle of binocular system target positioning is introduced.The commonly used visual coordinate system and its transformation matrix are listed.(2)A target tracking algorithm based on TLD framework is proposed.This algorithm expands the detector module in classical TLD algorithm by Adaboost algorithm and improves the speed of target detection.The algorithm uses Harris corner as the optical flow tracking point,and the optical flow calculation calculates the optical flow tracking point by combining the FB error and the MSD matching error,which improves the calculation precision and the algorithm efficiency.(3)Most of the UAV tracking systems in the market are only suitable for high-altitude environments and therefore can not track small targets.Since the size,shape,etc.of the target in the field of view change over time,these systems can not effectively track the target for a long time;Common UAV tracking system can not be separated from the GPS navigation system,while tracking the process can not guarantee the relative position with the target remains unchanged.In view of the above problems,this paper presents a navigation algorithm based on TLD frame tracking-SIFT feature point matching in target area-disparity calculation-target pose estimation-waypoint calculation.In view of whether the target shape is known or not,this paper proposes a method of target pose estimation for PNP/RANSAC algorithm.In order to solve the problem that the absolute coordinates of the waypoints can not be calculated when the GPS is out of lock,a method of calculating the edge orientation under the camera coordinate system is proposed,which realizes the guidance of pure targets without GPS.(4)This paper designs a variety of experimental scenarios,verify that the algorithm can be applied to the default scene and solve the problem in(3),in which the shape of the pedestrian changes and the scene of the vehicle tracking rapidly changes.In the indoor environment,the target shape of the known and unknown target attitude measurement.In the outdoor complex environment,we completed the experiments on pedestrians and vehicles and guided landing experiments at high altitude "overlooking" mode,completed the target tracking experiment at low altitude "head-up" mode and completed the target guidance experiment at low altitude without GPS Overcome the problem of sensor signal instability.