Research On Vehicle Detection Of Multi-lane For Traffic Intersection Based On Image

With the economy development of our country, the traffic pressure of medium-sized and large cities is increasing year after year. Intelligent Transport System (ITS) has been being an important way to improve the efficiency of transportation, playing a role in large cities in home and abroad.It is difficult to build a model of traffic net in city, then ITS can enhance its power with the support of a lot of timely traffic information. It cost much money and human source to get the traffic condition information by the classical physical sensor, large number of image/video device were installed above the road in most medium-sized and large cities, acquisition of timely traffic information through these device is hot research point, No extra hardware is needed by using them, but the enhancement of the software function, which refer to the image processing and recognition technology, is necessary. So image processing technology means a lot to the development of ITS. This paper present a method of vehicle detection based on image.This paper firstly describes the development of ITS and the research condition of vehicle detection, then, introduces the image processing technology:image enhancement、 image segmentation、corner detection、morphological processing of binary image, at last, a new method of corner detection and image segmentation is proposed, recognition of vehicles in an image by BP neural network is elaborated, too. The method of vehicle detection proposed can be used in the traffic intersection for vehicle detection and counting. The main contents are as follows:1、Vehicle detection. We get a clear gray-level image by the pretreatment of the image obtained by camera. Firstly, traversal the image with a new mask, removing the flat area according to the characteristic of the pixels’ variance, secondly, detect the corners with the edges, after getting the corners, the corners are adopted as the seeds of region-growing, finishing the region-growing, we get the binary image with several connected areas. 2、Vehicle recognition. To recognize the vehicles in the binary image, This paper gets the sub-area by template matching in the binary image, moving the template in the image, if the sub-area in the binary image is similar to the template, then, extract the sub-area, getting the small image with the pixels of "320X240", and, calculate the Invariant Moments and number of corners of the small image, put the feature vector consist of the Invariant Moments and number of corners into the BP neural network, it tells whether it’s a vehicle or not."1" for yes, and "0" for no. Finally, we get the number of vehicles in the big image.