Research On Vehicle Speed Measurement Technology Based On Road Marking Semantic Segmentation

In recent years,the automobile industry has developed rapidly,and the number of cars owned by residents has increased year by year.However,due to factors such as road infrastructure,the number of traffic accidents has remained high.For the handling of traffic accidents,most cases require vehicle speed measurement as an important basis for post-accident liability determination.The traditional vehicle speed measurement method mostly uses radar and other equipment to measure the speed of the vehicle,but for the vehicle speed measurement of accident identification,this kind of speed measurement is more troublesome.With the widespread use of video surveillance,the video resources of road traffic are more and more easily obtained,so the video-based vehicle speed measurement method is more and more widely used in vehicle speed measurement.At present,the vehicle speed measurement in accident identification mainly adopts manual manual measurement of vehicle speed,manually selects the reference object in the video to determine the driving distance of the vehicle,and then combines the speed measurement method to obtain the driving speed of the vehicle.This method is simpler than the vehicle speed measurement method using camera calibration,but this method of relying on manual speed measurement still has the problems of time-consuming,labor-intensive,low-efficiency and large manual influence.In view of the above problems,this paper proposes a vehicle speed measurement algorithm based on road marking semantic segmentation for vehicle speed measurement of in-vehicle video,which realizes the intelligent and automatic speed measurement,and implements the vehicle video speed measurement software platform based on this algorithm.The main research work and research results of this paper are as follows:First,this paper studies a road marking segmentation algorithm based on semantic segmentation network.By analyzing the characteristics of the road markings in the video,the Deeplabv3+ network is used to segment the road markings in the image.Aiming at the inaccurate segmentation of road markings by the original Deeplabv3+ network,three methods of weight balance,feature fusion and attention mechanism are proposed to improve the Deeplabv3+ network.The improved network improves the segmentation accuracy by22.2%,which is more suitable for the segmentation of road markings.Second,this paper proposes a speed measurement algorithm based on road marking semantic segmentation by combining the ideas of road marking semantic segmentation and traditional video speed measurement.By analyzing the road markings,using the lane dividing line as the reference for speed measurement,and aiming at the problem of incomplete segmentation of the lane dividing line,an optimization algorithm is proposed to redraw the segmented image.For the problem of how to automatically trigger the speed measurement,a fixed disappearing position of the lane dividing line in the process of the vehicle is proposed as the triggering condition of the speed measurement.Finally,it is verified by experiments that the average absolute error between this algorithm and the traditional speed measurement algorithm in the test video experiment does not exceed 2.38km/h,and in the real accident video speed measurement experiment,the error does not exceed 1.13km/h.Third,build a software platform based on in-vehicle video speed measurement,adopt B/S architecture,and develop and implement Python-based Flask architecture.According to the actual speed measurement requirements,corresponding functions are designed to visually display the video of the speed measurement results and the speed measurement curve,which is convenient for appraisers to analyze the overall driving process of the vehicle.