Visual-servo-based Ground Object Tracking For Four-rotor UAV

Ground Object Tracking and autonomous landing for UAV are hot topics in UAVrelated technology research.This project uses the drones which perform well on the tasks of surveillance,target guidance and autonomous landing in relatively complex and narrow small scenes,and studies the object tracking and autonomous landing technology of the drones.In order to realize stable object tracking for the drone,high requirements are placed on the performance of the object detection and tracking algorithm and the controller.This thesis mainly studies the following three problems: First,it solves the problem that the change of the appearance of the object during the movement will result in that the tracking result will drift or even the object detection will fail.Secondly,after obtaining the tracking results through the object detection and tracking algorithm proposed in this thesis,the appropriate control strategy is designed to achieve the stable object follwing.Thirdly,in order to facilitate the recycling of the drone,the autonomous landing of the drone is realized after the end of the mission.Firstly,the stable object detection and tracking algorithm is the premise of the UAV to achieve object tracking.This thesis proposes an object detection and tracking algorithm for the UAV platform.Using the advantage of off-line learning based on deeplearning-based object detection algorithm,SSD algorithm(Single Shot Detector)and Parallel Tracking and Detection(PTAD)are combined,overcoming the insufficiency of traditional detection algorithms on detecting the appearance changed object.Using the characteristics of the PTAD framework,the efficiency the detection module is improved,thereby effectively improving the tracking performance of the entire framework.Secondly,the suitable control strategy is the key for the UAV to achieve object tracking.The deviation of the position of the tracking result in the image from the expected position is used as the error signal for the control.The image-based visual servoing control method is used to design the PID controller to realize the stable tracking for the drone using the forward camera.Finally,in order to facilitate the recycling of the drone,this thesis designs a landing sign for the drone with a QR code in the circle.Through a fast and effective circle detection algorithm,the drone can searchthe landing area,and the image-based visual servoing control method is used to design the PID controller to realize autonomous landing.During the landing process,the drone determines whether the drone should land by identifying the QR code to avoid the false landing.In this thesis,the AR.Drone 2.0 from Parrot,France,is used as the experimental platform to verify the effectiveness of the proposed methods in this theisi.