Research On Target Tracking Through Uav Based On Fusion Of Vision And Depth Information

Unmanned Arial Vehicle(UAV)based target tracking is a significant research subject that can be widely used in military and civilian domain.In this paper,UAV target tracking approaches based on fusion of vision and depth information were proposed and a series of research work were carried out,consisting of system modeling,image processing,target recognition,target tracking and flight controller design.First of all,the problem of UAV target tracking based on vision and depth information fusion were analyzed.A systematic framework of target tracking control was proposed,in which,the key subproblems that need to be solved were identified.Subsequently,the kinematic and dynamic model of the quadrotor,camera model,target motion model and the kinematic model of the UAV target tracking system were established and further,based on reasonable assumptions,the dynamic model of UAV was simplified and linearized.Secondly,the SURF feature detection algorithm was studied for the problem of target recognition.The target recognition effect and the real-time performance were verified in four aspects: the target scale transformation,the target rotation,light intensity variation and the target occlusion.In necessity,noise filter was employed to the image with noise,which reduces the effect of noise on the recognition of the object.Thirdly,the Camshift algorithm and the Kernelized Correlation Filters algorithm were applied for the target tracking in the image domain and the KCF algorithm is improved in two aspects.On one hand,improvement via adjusting the search window by motion model based predicting the location of the target center solves the problem of tracking drift or loss due to fast and sharp movement.On the other hand,improvement via optimizing template update rate based on the judgment of the target occlusion makes the template adaptively update and solves the problem of tracking failure due to occlusion.For the target tracking in the spatial domain,the result of image field tracking was combined with the depth information provided by a Realsense camera,and the high precision estimation of the target three-dimensional position coordinates was realized.The tracking effect of Camshift algorithm was comepared with that of KCF algorithm.Then,based on the positional PID control algorithm and the dynamic characteristics of quadrotor,a UAV tracking controller with double closed-loop structure was designed.A Simulink model of the control system that is used for verifying the performance was established.The simulation results indicate that it has good stability,accuracy and rapidity.Finally,a UAV target tracking system experimental platform was built.The attitude control experiment and the target tracking experiment were conducted on the platform.The experimental results show that the ideal attitude control and target tracking can be realized on this system.