Research On Localization And Target Tracking Algorithms Of UAV Without GPS Signal

In recent years,unmanned aerial vehicles(UAVs)have attracted more and more attention due to their flexibility,high maneuverability,and low cost.After combining a variety of sensors,UAVs can be widely used in military and civilian fields.Moving target tracking of UAVs is the current research hotspot in the field of UAVs.However,when the UAV is working in a GPS-free environment,the tracking fails due to the UAV’s inability to locate itself,which limits the application scenarios of UAV target tracking.Therefore,in order to achieve the tracking of the UAV in the non-GPS environment,not only a real-time and robust target tracking algorithm is required,but also the estimation of the UAV’s pose is required.In response to the above problems,this paper independently designed and built a UAV target tracking system,carried out the research on the corresponding algorithm of UAV localization and target tracking,and achieved the autonomous tracking of ground moving targets in GPS-free environment.The specific work is as follows:(1)Aiming at the hexarotor,the corresponding coordinate transformation model is constructed and the UAV kinetic model is modeled.(2)Aiming at the localization problem of the UAV in the non-GPS environment,this paper proposes a localization and mapping algorithm based on the fusion of Inertial Measurement Unit(IMU)and stereo visual information to complete the localization of the UAV,which effectively improves the robustness of visual localization.Finally,experiments have proved the effectiveness of the algorithm.(3)Aiming at the image sequence tracking problem of UAV,a siamese network target tracking algorithm based on Mobile Net V2 is proposed,which can run in realtime on the UAV onboard processor.The algorithm uses Mobile Net V2 as the feature extraction network,including two parts: target score estimation module and target scale estimation module.Among them,the target score estimation module adopts an offset learning method for training,which can estimate the initial position of the target in current frame.The target scale estimation module can predict the Intersection over Union(IoU)between the target bounding box and the real bounding box,and use the gradient ascent method to iteratively correct the position and scale of the target.Then the multi-layer features of Mobile Net V2 are used and the residual fusion strategy is adopted to obtain the tracking output.Finally,experiments have proved the effectiveness of the algorithm.(4)Aiming at the problem of target loss caused by full occlusion in the UAV tracking process,this paper designs a re-detection algorithm based on saliency detection.This algorithm can efficiently predict the saliency map of the image in realtime to guide the re-detection of the target,and then resume tracking.Aiming at the tracking control of the target by the UAV,this paper designs a corresponding controller,which adopts the speed and position outer-inner loop PID controller to keep the target in the center of the UAV’s field of view.Finally,through simulation and actual UAV tracking experiments,the performance of the algorithm and the feasibility of the tracking system are proved.