Study On Method Of Vehicle Detection And Tracking Based On Monocular Vision

With the development of society, more and more people buy cars, and accordingly,traffic accidents are also increasing. In order to solve this problem, more and more countries have begun to study the intelligent transportation system. And the core of the basis of intelligent transportation system is to detect and track vehicles on the road, according to the position information of vehicle to avoid traffic accidents. This paper is based on the visual method for detection and tracking of vehicles.Vehicle detection is usually divided into two steps. Firstly confirm the vehicle may exist in the region, including the roadside trees left in the false. The next step is to remove the false shadow, confirming the specific position of the vehicle. Under normal circumstances, the vehicle in front the car collision most likely come in the lane. So this paper first detects lane to narrow the scope of vehicle detection, improving the efficiency and precision of detection. Based on the lane line, shades of gray and road vehicles on the road by the bottom of the value of the larger contrast, confirm the vehicle may exist in the region. Then fuse image entropy features such as texture to eliminate false vehicle shadow,and accurately detect the vehicle in front.The main work of this paper is as follows:1. Improving OTSU adaptive threshold method, estimate the road region gray value by taking the road statistical sample region gray value characteristics,which can avoid processing the whole image of the disadvantage of OTSU. Using this method also can save processing time, improve the real-time performance of the algorithm and accuracy.2. In the lane line detection algorithm, using the method of morphological and edge extraction, design an algorithm that can search for the inner edge of the lane line. By contrast with the performance of Hough transformation,this paper chooses the method of least squares to fit lane. In order to further improve the efficiency of the algorithm, this paper adopts a lane tracking algorithm,extended 50 pixels wide searching on each lane of a previous frame image position, greatly reducing the time of image processing. According to the lane line detected, this paper calculates the yaw angle of the vehicle every frame image.3. In the vehicle detection and tracking algorithm of this paper,based on the algorithm of shadow detection, eliminate the false vehicle region with image entropy and gray image symmetry, and detect the position of the vehicle in the image information. Then track vehicle based on the method of Calman filter, while guaranteeing the detection precision and improving the detection efficiency, enhance the real-time algorithm.4. This paper established the anti-collision model of safe distance, that is the relationship between the relative speed and maximum braking distance. And given the location model based on vision, according to the image coordinates can be detected in the vehicle calculates the distance, then estimate the time of collision.This paper designs a software of front vehicle detection system by using VC++language and the visual processing library named OpenCV1.0. The experimental results show that the algorithm meets the real-time requirements. And in the light conditions are good road can stable tracking of vehicles in front of goal, but also has certain robustness to complex road conditions.