Traget Tracking System Of Four-Rotor UAV Based On Machine Vision

Target detection and tracking technology in machine vision is a hot topic in current research.As the traditional target tracking systems usually adopts static video acquisition equipment,video acquisition range is limited.With the continuous maturity of unmanned aerial vehicle(UAV)technology and the reduction of related product prices,the application of machine vision technology to UAV is a new breakthrough,which provides new ideas for target tracking technology and attracting more and more researcher’s attention.Among them,the four-rotor UAV has the unique advantages of strong maneuverability,good concealment and high flexibility.This paper uses the four-rotor UAV as the carrier to carry out related research on the target tracking technology of the UAV.Aiming at the shortcomings of the traditional target tracking technology,this paper proposes a solution of target tracking based on the combination of PTZ and UAV.The main research work of this paper is as follows:(1)Firstly,in order to build a four-rotor UAV target tracking hardware system,the overall design structure of the UAV target tracking system was planned,and the hardware system was put forward.The hardware system is divided into a flight control system and visual tracking system with the modular method.The system hardware solution is also designed and selected.(2)Arrording to the software design problem of the four-rotor target tracking system,the UAV position and attitude controller and the PTZ controller based on PID control are designed by analyzing the flight control principle of the four-rotor UAV and the control principle of the PTZ.The machine visual image part adopts the target detection method combining FAST and SURF to design a visual information transmission protocol.(3)In view of the problem that the target tracking system may encounter occlusion or target scale change,the commonly used mainstream target tracking algorithms TLD(Tracking-Learning-Detection)and KCF(Kernel Correlation Filterare)deeply studied,and on this basis,the KCF algorithm is improved in occlusion and scale change.Through the comparison experiments in the OTB dataset,the accuracy of the improved KCF algorithm is 4.6% and 21.4% higher than that of the KCF and TLD algorithms,and the success rate is increased by 30.8% and 9.8%respectively.(4)Based on the above hardware system,software system and target detection and tracking algorithm,the experimental platform of the UAV target detection and tracking system was tested,includeing the pan-tilt tracking experiment based on vision,four-rotor fixed-height mode flight experiment,four-rotor target detection and tracking experiment.However,the current target of tracking is mainly based on the target on the ground such as people and vehicles.This paper attempts to use the airborne drone target as the research object to complete its target detection and tracking tasks.