Research On Vehicle Detection And Tracking Based On UAV Platform

Using the unmanned aerial vehicle to shot the expressway can achieve a wide range of traffic data acquisition flexibly,timely response to sudden traffic accidents and congested road sections,and convenient traffic management departments to monitor and control,with broad application prospects and great economic value.Therefore,this paper studies the vehicle detection and tracking algorithms under the unmanned aerial vehicle platform.The research includes the following aspects:(1)Research on vehicle detection algorithms under UAV platform.Based on the analysis and improvement of FaceBoxes network model,a real-time vehicle detection method on UAV platform is designed.Firstly,the network structure is redesigned based on the multi-feature fusion idea,so that the vehicle prediction can contain more semantic information and feature details.Then,multiscale anchor design is carried out to improve the adaptability to different vehicle sizes and small target vehicles.Finally,the algorithm is time optimized based on the binary weight network.Experiments on the aerial vehicle datasets show that the proposed algorithm has better vehicle detection accuracy and real-time performance under the UAV platform.(2)Research on vehicle tracking algorithms under UAV platform.Based on the vehicle detection results,a kernel correlation filtering based vehicle tracking algorithm is designed for the analysis of kernel correlation filtering algorithms.Firstly,based on the KCF algorithm,the state estimation model of the vehicle is established,and the state estimation reliability judgment method is proposed.When the kernel correlation filter state estimation is unreliable,the Vehicle motion information is used to estimate the vehicle state.Secondly,data association between vehicle status estimation results and vehicle detection results of improved FaceBoxes network is carried out,and a multi-level data association strategy is proposed.Finally,the vehicle motion trajectory update strategy is designed,and the vehicle motion trajectory and state estimation model are updated based on the data association results.Through comparative experiments on aerial vehicle datasets,it is shown that the proposed algorithm can be applied in real-time under UAV platform and has good vehicle tracking accuracy.(3)Software design and Implementation Based on UAV platform.Vehicle detection and tracking system is developed based on UAV platform,which realizes real-time vehicle detection and tracking for aerial video by airborne equipment,and has the function of vehicle flow density and traffic flow detection.At the same time,the mobile software can acquire the information of monitoring road vehicle movement and control the operation of airborne software.