Research On Vehicle Queue Length Detection Method Based On Image Processing

With the gradual improvement of the national economy,the increase in the number ofprivate cars has brought great convenience to people.However,social problems such as traffic congestion have emerged.The increase of congestion in traffic scenarios presents challenges to Intelligent Transportation System(ITS).The vehicle queuing length detection based on image processing is an important concept in the intelligent traffic management system.It can produce queuing information on the road immediately and accurately,which is of great significance for alleviating traffic congestion.In this paper,the traditional image processing method and the image processing method based on the deep learning have been used to the problem of vehicle queue length detection,with the data of video and picture.(1)A queuing length detection method based on the feature point tracking has been constructed.Firstly,the single motor vehicle lane is separated as an interest area by automatically detecting the lane line,and then the feature points are tracked in the interest area,and the track information has been used to determine the queue length.Finally,according to a simple camera calibration algorithm,the pixel distance is mapped to the actual spatial distance,thus obtain the actual queue length.It has been proved by experiments that the detection accuracy of F1-Measure is 91.96%,and the average error within 150 meters is5.17%.(2)A queuing length detection method based on deep learning object detection has been constructed.By analysing the instability problem of the inaccurate feature point detection and tracking in the unstable method,the vehicle object detection is carried out on the SSD framework which is based on the VGG16 network model,which has better detection performance than before.Then,the KCF algorithm has been used to track the vehicle object to obtain its motion state.Finally,the vehicle queue length is calculated according to the position of the object detection,and the distance of the queue length for the large vehicle has been compensated.Comparative experiments show that the performance of the deep learning methods has been improved.The experiments show that the F1-Measure is increased by3.76% compared with the traditional method,the average error is reduced by 1.33% within150 meters,and the number of vehicles in the queue can be obtained.(3)A queuing length detection method based on the image segmentation based on deep learning algorithm is constructed.For the object occlusion problem in vehicle object detection,the image segmentation algorithm based on deep learning has been selected.By strictly labeling the data set,the full convolution network which is modified by VGG16 is used to segment the vehicle queue.Then the length of the pixel marked as a queue in the model detection result has been calculated,and the actual queue length is obtained according to the camera calibration.The experimental results show that the detection accuracy of F1-Measure is 97.53%,and the average error within 150 meters is 3.06%.Compared with the first two video-based methods,the image segmentation algorithm is based on a single image,which requires less hardware and memory and has greater application value.