The Algorithm Research Of Vehicle Flow Measuring Based On Video Image

With the sharp increasing of the vehicle, many traffic jams have occurred and thecost of traffic management has been increasing dramatically. One of the mosteffective ways to solve these problems is to gather the traffic parameters effectivelyand distribute the resources of road reasonably. The algorithm research of vehicleflow measuring based on video image is one of the ways to gather traffic parameters，using communication technique and computer technique and so on. It has manyadvantages, such as rich information, low cost and universal performance. It is animportant part of intelligent traffic system and has a broad prospect.In the process of traffic statistics, the integrity of moving objects is important tothe accuracy of vehicle flow measuring. In this thesis, the background difference andimproved two adjacent frames difference are adopted to detect moving objects. Theimproved two adjacent frames difference is used to detect the moving object, which isnot be detected by background difference. What’s more, the ideal background is alsoimportant to the detection of moving objects. This thesis proposes a mean algorithmbased on composite background images. It calculates the average twice. Firstly itcalculates the average of frames and the initial background comes out. The initialbackground is updated by Kalman filtering. Secondly it calculates the average ofcomposite background images and the ideal background comes out. The compositebackground image is a frame whose area of moving objects is replace by the renewedbackground. The ideal background needs to update in order to adapt to the changes ofweather and illumination. This thesis proposes an improved method of backgroundupdate based on statistics. It calculates a new background based on statistics.However, the shadow of moving objects has a bad effect on the statistics of adjacentroadway. This thesis uses the method of normalized cross correlation combined withgray value detection and value detection in HSV model. This method can detect thewhole shadow and avoid the interference in traffic statistics of adjacent roadway.This thesis uses virtual-loop sensor to count the vehicle. The virtual-loop sensoris setting along the vertical direction of the road. The data in the virtual-loops sensorarea is processed in order to reduce the time cost of calculation and improve theprocessing speed. The experimental results show that the algorithm in the thesis isefficient and can count the vehicle accurately.