Autonomous Target Tracking For Multi-rotor UAVs Based On KCF-TLD

The multi-rotor UAV is an unmanned aerial vehicle with good flying performance,such as hovering,flexible movement,compact mechanical structure,and high system reliability,its attitude is adjusted by changing the rotation speed to achieve position control.It is easier to obtain current environmental information through the sensors mounted on the UAV,and then issue control instructions to guide the vehicle to complete specific flight tasks.In this thesis,a visual sensor is selected to obtain video stream of the current environment and the tracking targets,then an environment adaptive tracking algorithm is studied to control the vehicle to achieve the goal of autonomous target tracking.The main contents are as follows:To study the autonomous control of UAVs,the mathematical models are established for autonomous target tracking system based on the multi-rotor UAV,including the UAV's own dynamic model and the visual target imaging model.The UAV dynamic model is to analyze the UAV flight attitude control principle,and the target imaging model is to compute the relative position relationship between the UAV and the target,and furtherlay the foundation for UAV autonomous target tracking.In visual tracking part,the principle and mathematical model of KCF(kernel correlation filtering)algorithm and TLD(Tracking-Learning-Detection)algorithm are studied at first.Then based on the image sequence obtained by the UAV's mounted camera,the fused KCF-TLD algorithm is studied in this thesis,which introduces the KCF algorithm to replace the tracking module in the original TLD algorithm.After testing,it is found that the KCF-TLD algorithm shows good real-time performance and performs well on target occlusion,but in.the case of insufficient ambient brightness,the target may not be recognized.Therefore,a brightness enhancement module is proposed in the KCF-TLD algorithm to ensure accurate target recognition.Next,for the established UAV model,a cascade PID controller is designed and simulated in MATLAB to verify the validity of the model.To verify the effectiveness of the tracking algorithm,the algorithms is implemented on the actual platform.Through the monitor on PC,we select target manually,and the commands are sent to control the vehicle to adjust the posture and finally complete the autonomous target tracking task.