Algorithm Research And Implementation Of The Video-based Moving Vehicle Detection System

Vehicle detection and counting is one of the most important contents of the researchof Intelligent Transportation System (ITS).Based on the analysis of the existing vehicledetection technology, this paper studied the algorithms about moving target detectionbased on digital images and video-based vehicle detection. Then, this paper designed andimplemented a video-based vehicle detection system. Existed vehicle detectiontechnologies often adopt just a single method to count the traffic flows, which makethemselves hard to be applied to more complicated scenarios. Given the above reason,this paper proposes a method that integrates the detection line technology to detectnormal traffic flows and the virtual loop technology, when traffic jams happen, toimprove the accuracy of the statistics of the traffic flows. The paper involves thefollowing parts: moving vehicle detection, vehicle shadow elimination, vehicle lightingelimination, traffic jam detection and the shift of different detecting methods, which willbe sketched below.(1) vehicle detection researchBased on the study of existed vehicle detection algorithms, the paper adopts themost widely used background subtraction method. What’s more, the paper proposes apartition piece background update method based on the accumulative multi framedifferential of the sampling points. Compared with the traditional methods, our methodcan improve the speed of the computing and the update speed of the background greatlywith small calculation consumption and ensure the real-time performance of the system.(2) vehicle shadow eliminationThe shadow moves with the vehicle, which belongs to the foreground. If noteliminated or weaken, the shadow will be detected as a moving object to disturb thesystem with wrong information. This paper adopts a common elimination method basedon HSV color space.(3) vehicle lighting eliminationAt night, the lamplight of the vehicle will produce a light spot with high intensity on the road in front of the vehicle, which will make trouble with the system if not handledproperly. Based on many experiments, the paper proposes an algorithm of graduatedsymbols to eliminate the light of the vehicle. The method has achieved desired results.(4) shift of traffic jam detection and counting methodAs our method includes two means of detecting and counting, the system can workwell when the traffic is either normal or bad. When the traffic flows normally, the systemuses background subtraction method to detect the vehicles and counting them based ondetecting line technology. Detecting line technology makes sure the result is right whenvehicles change their lanes and the real-time background update help improve theaccuracy of the moving objects extraction. When the traffic jams happen, system willterminate the update process of the background and adopts the spare background imageinstead to avoid the foreground being mixed into the background. And the countingmethod will be shifted to the virtual loop technology automatically to avoid the wrongextraction of adhesive targets. The shift between the two traffic occasions has been solvedin the paper to make the system robust and reliable. Besides, as to make sure the systemruns accurately, the paper proposes a jam detection method based on the mean change ofthe multi frame differential to help diagnose the real-time traffic status.After conducting a theoretical study of moving object detection, this paper designsand implements video-based vehicle detection and traffic counting system. By erectingcameras at important intersections, we use image processing techniques to detect themoving vehicle targets. As an important part of the Intelligent Transportation System ofWuhan City, it can automatically detect the main road bridge traffic junctions, and vehicledetection accuracy rate of more than99%. The system can improve the efficiency of thetransport system in Wuhan City.