Research And Implementation Of Quadrotor UAV Formation Based On Visual Tracking

Unmanned aerial vehicle(UAV)is widely used because of its compact,flexible and easy to operate features.Due to the limitation of single UAV’s endurance and load,the formation problem of UAV has received wide attention.However,most UAV formation flights are based on inter-aircraft communication.If the communication link fails or is interfered by external signals,the task will be affected.For this reason,this paper suggests that visual tracking based formation can be used to improve these shortcomings.In this paper,the distributed Leader-follower formation mode,combined with the robot operating system and deep learning algorithm,and the use of visual information navigation are important means of formation flying in complex situations.This paper focuses on the in-depth study of some key issues of the quadrotor UAV platform,the main work is as follows:1.Design the software and hardware for the overall scheme of the platform according to the actual needs,and select appropriate power system,onboard computer,rack and other components to build the platform.2.Build the related dataset in this paper,use the dataset in the Keras framework to train the YOLOv3 algorithm and the improved Mobile Net-YOLOv3 algorithm,and compare the related performance indicators before and after the improvement,finally deploy the lightweight model in the onboard computer,and use Tensor RT to optimize it.3.Use the improved algorithm to get the center point coordinates of the leader machine,get the position of the leader machine in the absolute coordinate system through the conversion rules between the related coordinate systems,and move the follower machine to the specified position through the corresponding control algorithm to complete the whole visual tracking process.4.Conduct real flight tests on the formation,and use MATLAB tools to analyze the exported flight data to verify the stability of the formation system.