Research On Re-Identification And Tracking Of Moving Vehicles In Dense Traffic Scenarios Based On Mixed Features

Vehicle re-identification and tracking is the key point and difficulty of the research on intelligent transportation system(ITS).The frequency of dense traffic scenarios in urban road traffic systems is also increasing.Improving tracking accuracy in dense traffic scenarios plays a vital role in improving the performance of vehicle tracking systems.However,in this scenario,there are many challenges,such as small target,cross-camera vehicle re-identification,and identify switches(ID switches)caused by occlusion.In view of the above problems,this paper proposes a vehicle re-identification and tracking system based on mixed features to further improve the reliability and accuracy of vehicle tracking.This paper proposes a hierarchical recursive super-resolution reconstruction network,to realize the reconstruction of low-resolution vehicle images into high-resolution images.Hierarchical recursive module improves the detail reconstruction effect and shares parameters recursively,reducing the amount of parameters.The experimental results show that the hierarchical recursive network proposed in this paper can achieve a better super-resolution reconstruction of vehicle images.Aiming at the problem of cross-camera vehicle re-identification,this paper proposes a recursive residual network as a feature extractor to extract vehicle image features.In the recursive residual network,this paper proposes global residual learning and local residual learning to reduce training difficulty and transfer image details to the deeper network.Besides,in this paper the separation margin loss is used instead of the traditional margin loss,and the supervision training is based on the multi-loss supervision.Experiments show that the recursive residual network proposed in this paper can effectively reduce the impact of cameras on vehicle re-identification.Aiming at the occlusion problem in dense traffic scenarios,this paper uses a trajectory matching strategy based on mixed features to reduce ID switches.The mixed features include the appearance features and deep learning features of the target.Appearance features consist of RGB,HSV,Lab,HOG,and LBP features.Deep learning features are the high-frequency features of the target obtained by the recursive residual network.The experiments on the vehicle tracking dataset show that the reconstruction of low-resolution images based on hierarchical recursive networks can reduce the impact of low-resolution images on feature extraction in dense traffic scenarios.At the same time,experiments also show that the vehicle identification and matching algorithm based on mixed features in this paper can further improve the accuracy of vehicle tracking and reduce ID switches.