Research Of Lane Recognition Technology In Fog Environment Based On Machine Vision

With the increasing need of automobile intelligence, the driving safety of vehicles in the severe weather is imperative to be addressed. The lane recognition technology based on machine vision as a critical part of car security aid system, is a effective solution to the driving in severe environment. The present work mainly focuses on the driving in normal weather, but driving in the severe weather, for instance, fog and haze, has not been thorough investigated. Therefore, with the problems of the low contrast of road image and the difficulty of lane identification in fog environment, the paper proposes a lane recognition method based on improved dark channel prior algorithm. The main contents are as follows:1. The background and significance of lane recognition method in the fog environment based on machine vision is given. Then, the lane departure warning system, lane recognition method and lane tracking methods are reviewed. Subsequently, several existing problems on lane detection are discussed. And the main work of this thesis is also demonstrated.2. For low contrast of road image that is acquired by vehicle mounted camera in the fog environment, the paper introduces a concept of image defogging in image preprocessing stage. By comparing effects of different defogging algorithm on the fog road image, this paper proposes an improved dark channel prior algorithm, which can not only meet the real-time, but also can achieve good defogging effect and enhance the contrast of the road image. When processing the fog road image by defogging algorithm, the pixel gray value of the image above 1/3 is set to 0. If using traditional global OTSU method to segment the road image which has been handled, segmentation threshold that is obtained will be too small and binary image will has too much noise. So the paper uses local OTSU method for image segmentation.3. To detect bend lane, this paper adopt straight- curve model. The road image is divided into two parts called near area and far area. The straight line called neighboring area is detected by using the improved Hough transform, and the curve line called far area is fitted by least square method. For extracting curve feature points of far area, we firstly determine vanishing point and detecting area of curve feature points based on straight line parameters. Because of the impact with other interference factors(such as noise and so on) of the road, we probably extract invalid points or other useless feature points. So we must determine the credibility of curve feature points. When using least squares method to fit curve, we confirm cubic curve fitting to the best by comparison and analysis.4. For the single frame road image of video sequence, it will reduce the real-time performance of the system repeatedly to detect lane. According to the small difference of the information between the adjacent continuous two frame images, we use Kalman filter to track lane. Therefore, it improves the real time and accuracy.5. In order to verify effectiveness of the lane detection method in the paper, we analyze and verify the algorithm by simulation and experiment in MATLAB. The results show that this method can not only meet the real-time requirements, but also can accurately recognize the lane in the fog environment. At the same time, the improved algorithm is less than 10% of that of the dark channel prior algorithm.