Research And Design Of UAV Tracking System For Non-Cooperative Target

Quadrotor UAVs are widely used in aerial photography,rescue,military and other fields due to their convenient take-off and landing,strong maneuverability,and simple control.Among them,the vision-based UAV target tracking system has become a hot direction of the UAV,which can be done only through the airborne camera.The detection of targets and environments can be achieved at a lower cost for tracking non-cooperative targets in unknown environments.This thesis elaborates on the research and design of UAV tracking system for non-cooperative targets.The specific content is as follows:Aiming at the problem of target detection and positioning in the tracking,in order to obtain reliable target detection results and ensure the detection speed,the Tiny-YOLOv3 algorithm is used for target detection to obtain the position of the target in the image,and Tensor RT is introduced to optimize the detection network.It is proved through experiments that the detection speed of the adopted method is much faster than that of YOLOv3 and Tiny-YOLOv3.In order to obtain the target position,the disparity map is obtained by the SGBM stereo matching algorithm,then the depth of the target is calculated,and the target position can be solved.Finally,the reliability of the method used in solving the target position is verified through experiments.Since the tracking environment is unknown,the obstacle information acquisition method based on visual point cloud is adopted.On the basis of the disparity map obtained by the SGBM algorithm,a dense point cloud can be obtained.For convenient storage and use,the octomap is used to sparse the dense point cloud,and the accuracy of the obtained point cloud data is verified through experiments.In order to solve the problem that the target cannot be detected due to occlusion by obstacles,the Kalman position prediction method based on the target motion model is used.When the target is lost,the future position of the target will be predicted.Simulations verify the effectiveness of the prediction method.For the trajectory planning in the target tracking,based on the rapid expansion of random tree(RRT)algorithm,an improved RRT algorithm is proposed to optimize the selection of random points.If the trajectory does not pass through obstacles when the target point is used as a random point for expansion,then the random point is the target point,otherwise it is randomly selected from the given random points,so the planning efficiency can be improved.At the same time,the calculation of the yaw angle is added to the planning process,so that the UAV can keep the target in the center of the camera view during the flight.In order to solve the tortuous of the trajectory produced by the RRT algorithm,a redundant waypoint tailoring strategy based on the Bresenham algorithm is proposed.Finally,the effectiveness of the planning method is verified by simulation.In order to ensure the safety of the experiment,the Gazebo software was used to set up a virtual simulation environment,and the effectiveness of the method proposed in this article was verified through simulation experiments.And through the hardware selection,construction and software system design,the actual tracking system was finished.Finally,through actual flight experiments,the reliability of the UAV target tracking system proposed in this thesis was verified.