Research On Target Detection And Tracking Technology Based On UAV Perspective

Rotary-wing UAVs are small,hoverable,highly stable and easy to control.With the rapid development of embedded technology,more and more functions can be realised with rotary-wing-based UAV platforms.In particular,the combination of UAV platforms and vision modules enables UAVs to detect and track targets autonomously.However,there are still many aspects to be improved in order to achieve accurate target recognition and positioning from the perspective of UAV,and to tracking them stably.In terms of target detection,general detection networks are time-consuming and difficult to achieve real-time detection on limited performance airborne processing equipment,At the same time,the target scale from the perspective of UAV is small,which is prone to problems such as missed detection and false detection of targets when performing detection tasks.In terms of target tracking,the target is easily affected by background and occlusion,and stable tracking of the target over a long period of time remains a problem to be solved.In order to solve the practical problems that UAVs may encounter when performing detection and tracking tasks,this paper designs a UAV target detection and tracking system based on deep learning and correlation filters,with the following main work:To address the problems of time-consuming and low accuracy of UAV target detection,an attention mechanism-based target detection algorithm is proposed with the YOLOX-Nano algorithm as the base algorithm.Firstly,the SE attention mechanism module is embedded on the backbone network to focus on the more important information in the channel.Secondly,based on the Transformer structure,a feature fusion module based on the cross-attention mechanism is designed to fuse the backbone network output features with the Pan＿out2 features of the neck to enrich the global feature information of the target context.At the same time,the GIo U loss function is introduced to improve the convergence speed.Finally,experiments are conducted on the Vis Drone dataset,and the results show that this algorithm can improve the accuracy of UAV target detection on the basis of guaranteed detection speed.Aiming at the problem of background interference and occlusion in UAV target tracking,a target tracking algorithm based on feature fusion and adaptive update of the model with a priori mechanism is proposed by improving on the basis of the kernel correlation filters algorithm.Firstly,the color name features are adaptively fused with the gradient histogram features by setting the weight values based on the a priori information of each frame.Secondly,the APCE values and peaks are used to determine the degree of target occlusion,appropriate thresholds are set according to the degree of occlusion,and the appearance model is updated using different learning rates for different threshold cases.Finally,experiments are conducted on the OTB dataset to verify that the algorithm in this paper can accurately track targets in the presence of background interference and occlusion.The design of the quadrotor UAV target detection and tracking system,the construction of the hardware platform,and the target detection and tracking related experiments,the results show that the design of the target detection and tracking system has better robustness and faster operation speed,and can meet the practical application.