Design Of Vision Tracking System For Quadrotor UAV Based On Improved KCF Algorithm

Quadrotor UAV is an excellent aircraft with high degree of freedom,so it has strong maneuverability and can realize independent control of the x-axis and y-axis under the body coordinate system.It can be fixed-point suspension.Due to its excellent performance,the application field is extremely wide.This dissertation studies the design of visual tracking system based on four-rotor drones,including visual tracking algorithm and position control algorithm.Finally,the design of unmanned aerial vehicle and ground station is carried out on the physical platform.Firstly,the modeling work of the quadrotor UAV-based visual tracking system is carried out,including the establishment of the dynamic model and the target imaging model of the quadrotor UAV,and the establishment of the drone dynamics model as the position control algorithm.To lay the foundation,the purpose of establishing the target imaging model is to resolve the positional difference between the drone and the target.Secondly,based on the image sequence picked up by the UAV(including the object and its surrounding background information),an improved KCF(kernel correlation filtering)target tracking algorithm based on Kalman filter is used.The test result of the KCF target tracking algorithm on the test set shows that it has the advantages of antipartial blocking,light variation and other interference.However,after testing,it is found that KCF has many shortcomings in dealing with the rapid movement of objects or the challenges of camera shake during aerial photography.Therefore,this article adds prediction and correction functions based on the KCF algorithm.Subsequently,cascaded PD controllers and PD controllers were designed for the under-driven and full-drive subsystems of the drone model.Simulation results show that the designed tracking controller performs well.Finally,the algorithm implementation based on Opencv vision library and DJI SDK is implemented on the physical platform,including the implementation of visual tracking algorithm and position control algorithm of the UAV platform,and the tracking results are transmitted back to the ground end in real time through WIFI.The display module displays the tracking screen in real time for monitoring and can send mission instructions to the drone,such as take-off,landing,selection of the target object.