Research On High Speed Autonomous Landing For UAV

In recent years,with the rapid development of computer and artificial intelligence technology,the development and application of unmanned aerial vehicle(UAV)has been boosted.Quad-rotor UAV has been widely used in military and civil fields due to its compact structure,fast speed and convenience,low cost,capability of vertical takeoff and landing,easy control of flight attitude and so on.Combined with intelligent transportation,power line patrol,agricultural plant protection,military tracking,disaster rescue,aerial photography and other industry applications,to complete a variety of complex intelligent task,UAV will play an increasingly important role in the future.It will have important application value,to realize the recognition and tracking of moving targets and the autonomous landing of UAV on the moving platform.The autonomous landing technology is an key technology in UAV’s intelligent processing system,which includes three parts:(1)target detection and recognition;(2)target tracking;(3)position estimation and landing.Detection and recognition is the premise of UAV’s tracking and antonomous landing.In this paper,an improved Apriltags detection algorithm based on local searching strategy is proposed,which can greatly improve the processing performance and solve the problem of real-time performance.Thus,the stable detection and recognition of the target feature in the large scale changes are realized,and the target 3D position and attitude can be calculated in real time.Firstly,The mixed Gauss background model is used to detect the moving target,and then the classification is done by Faster RCNN based on Caffe.In the target tracking process,Kalman Filtering is used to estimate the target’s position in the 2D image,fitting function is used to predict the moving object trajectory,and the PID algorithm is used to realize the stable tracking.TLD algorithm can track a single target for a long time,and has the ability to track the target again after missing.And this thesis realized the stable recognition and tracking for face and moving vehicle based on TLD.In the autonomous landing process,through the specific landing Apriltags design combined with 16h5 and 36h11 Apriltags,we use redundant Apriltags information to calculate the target’s position and atittude.Besides,we use double closed loop PID control algorithm to ensure the rapidity,stability and accuracy of tracking and landing process,and finally realizes UAV’s autonomous landing task.The coordinate system is built for the whole system.And the designed algorithm is simulated and tested on DJI Matrice 100 programmable Quad-rotor UAV platform,equipped with Manifold,with Zenmuse X3 camera,with the mobile phone terminal APP DJI GO and UAV Simulation Platform DJI Assistant.The experimental results show that the detection,recognition,tracking and landing system designed in this paper is reliable.The algorithm realizes the UAV’s stable tracking and autonomous landing for fast moving target.