Vehicle Congestion Detection And Status Evaluation

Traffic environment perception,which can provide data support for traffic operation and management,is an essential part of the traffic operation system.Vehicle congestion detection is an important part of traffic environment perception and can identify congested areas in a timely manner.According to the situation of these congested areas,the departments of traffic operation management can formulate feasible solutions to unblock vehicle congestion.In this paper,detection and state evaluation of vehicle congestion are studied,mainly including vehicle detection,vehicle tracking and vehicle congestion state modeling and evaluation.In the part of vehicle target detection,the EM＿FCOS target detection network based on the FCOS model and the feature enhance module are proposed.The EM＿FCOS network uses the VGG-16 network as the backbone network.The feature maps output from the Conv3-3 and Conv4-3 layers of the VGG-16 network which are enhanced in EM＿FCOS are used to build the feature pyramid along with the feature maps output from the Conv5-3,Conv6 and Conv7 layers of the VGG-16 network.In addition,the method of pixel-by-pixel regression is applied in the feature pyramid.In the part of vehicle target tracking,based on the detection results of vehicle targets,the SORT algorithm is implemented in this paper to track vehicle targets.The SORT algorithm uses the Kalman filter algorithm to predict the position of the vehicle target in the next image frame,and matches the prediction with the target detection result in the next image frame by the Hungarian algorithm to accomplish the mission of vehicle target tracking.In the part of vehicle congestion state modeling,traffic parameters is extracted based on the results of vehicle detection and vehicle tracking,and the evaluation model of vehicle congestion state is established by the characteristic parameters such as vehicle density and vehicle speed.EM＿FCOS network can improve the detection accuracy of small-sized vehicle targets,and achieves 93.39%,90.75% and 95.47% accuracy of vehicle target detection on UA-DETRAC dataset,Pascal VOC 2007 dataset and MVD dataset,respectively.The SORT algorithm implemented in this paper achieves 92.61% accuracy in the MVT test set.The vehicle congestion state is classified by the vehicle congestion state evaluation model into smooth state,light congestion state and serious congestion state.The vehicle congestion detection and state evaluation system with human-computer interaction is designed and implemented by Py Qt5 framework,which can realize the evaluation of vehicle congestion state.