Research On Cross-camera Vehicle Tracking System In Expressway Tunnel

With the continuous improvement of China’s economic strength,the construction of expressways has developed rapidly,and the total mileage of expressways ranks first in the world.Due to the mountainous terrain in western China,a large number of tunnels exist on expressways.The tunnels are relatively closed and the environment is complex.They are the bottleneck sections of traffic and accident-prone sections.Related work is very important.Due to the limitation of the vehicle tracking range under a single camera,it cannot meet the demand for continuous tracking of vehicles in the tunnel.Therefore,it is necessary to study the cross-camera vehicle tracking system of highway tunnels.This thesis studies vehicle detection,multi-target tracking and cross-camera tracking methods in highway tunnel scenarios.The main contents are as follows:(1)Research on vehicle detection and tracking related algorithms.The traditional inter-frame difference method and the mixed Gaussian background modeling method and the YOLOv3 target detection model based on deep learning technology are studied,and the vehicle detection comparison experiment is carried out.The experimental results show that the detection effect of the YOLOv3 detection model is better in tunnel scenarios;Then the principles of Cam Shift and DSST tracking algorithms are introduced and the tracking effects of the two algorithms are compared through experiments.The experimental results show that the tracking effect of the DSST algorithm is better.(2)Research on multi-target tracking method under a single camera.A multi-target tracking method based on DSST algorithm is studied,which uses YOLOv3 as vehicle detector to initialize the DSST tracker.First,the principle of multi-target matching based on the Hungarian algorithm is studied,and then the new target determination,correction tracker and target departure determination in the multi-target tracking process are analyzed.Experimental results show that the method has good multi-target tracking effect.(3)Research on cross-camera vehicle tracking methods.Aiming at the internal environment of the tunnel,a vehicle tracking method for a cross-camera of a highway tunnel based on the space-time relationship is proposed.Firstly,perform a perspective transformation on the tunnel lane under the monitoring screen to determine the lane where the vehicle is located,and then perform cross-camera target matching when the vehicle passes through the overlapping field of view of adjacent cameras to complete the vehicle tracking and handover.Experimental results show that the method has good cross-camera vehicle tracking effect.(4)Design and experiment of cross-camera vehicle tracking system.On the basis of the above research,a vehicle tracking system for a cross-camera of a highway tunnel is designed and implemented.In the actual tunnel scenario,the vehicle data set is used to train the YOLOv3 model,and the trained YOLOv3 model is used to perform vehicle detection experiments on cars,passenger cars,and trucks in the tunnel.The experimental results show that the vehicle detection accuracy of the model reaches 97.25%;The actual tunnel video was used to conduct a cross-camera vehicle tracking experiment on the system.The experimental results show that the system’s cross-camera vehicle tracking accuracy is 96.72%.