Research On Vehicle Tracking Based On Deep Learning And Kernel Correlation Filtering

The increasing trend of the number of vehicles is very fast,which promotes the construction and development of transportation infrastructure.In order to improve traffic efficiency and facilitate people’s travel,relevant personnel focus on the study of intelligent traffic monitoring system.The importance of intelligent traffic monitoring system is becoming more and more prominent,and vehicle tracking is an essential part of it,so it is of great significance to study vehicle tracking.Aiming at various problems of vehicle tracking,a vehicle tracking algorithm based on deep learning and kernel correlation filtering is proposed in this thesis.The improved YOLOv3 network is used to detect the position of the vehicle,and the tracking frame of the kernel correlation filter is adjusted continuously according to the update strategy during the tracking process,so as to solve the problem that the kernel correlation filter is easy to lose the target vehicle in the tracking process,and improve the robustness of the algorithm to the scale change.High confidence update and APCE algorithm are introduced to solve the vehicle occlusion problem,and the updating strategy is improved.The algorithm proposed in this thesis is used to verify the data set.The results show that the proposed algorithm has a significant improvement in performance compared with other algorithms.The accuracy rate and recall rate of tracking target reach 90.2% and 92.4% respectively,which can meet the real-time requirements.Firstly,the network structure of YOLOv3 is studied.In order to improve the detection performance of YOLOv3 network,a layer of scale network structure is added based on it.Modify the network structure res2 to res4.In order to verify the performance of the algorithm,a large number of vehicle images were collected online and a series of preprocessing operations were carried out.Combining some vehicle images from the DETRAC data set as the experimental data set,the K-means algorithm was used to determine some parameters.The YOLOv3 network and Faster-RCNN network were compared with the improved YOLOv3 network.The experimental results show that the improved YOLOv3 verifies the superiority of network in both m AP and FPS.Secondly,the kernel correlation filtering algorithm is studied.The experimental results show that the kernel correlation filter algorithm can track the target in a short time,but the target frame will drift in a long time,and the performance is not good when the scale changes and vehicle occlusion problems are encountered.Due to the fast speed of the kernel correlation filtering algorithm,the improved YOLOv3 and kernel correlation filtering algorithm were proposed to be used for vehicle tracking,and the overall flow chart of the experiment was designed.The position of the detected vehicle is output from the algorithm,and the real-time speed of the vehicle can be roughly obtained after calculation.Finally,the labeling tool Labelimg is used to label the information of the vehicle images,and the verification experiment is carried out on the data set.Experimental results show that the proposed algorithm can not only solve the problem that kernel correlation filtering is easy to lose the target box in the tracking process,but also deal with the vehicle scale change.High confidence update and APCE algorithm are introduced to solve the occlusion problem of vehicles.The real-time performance meets the actual requirements,and the performance is greatly improved compared with other algorithms.