Research On Lane Detection Method With Three Front Cameras

Nowadays,due to the rapid development of information and computer technology,more and more scholars tend to rely on high-tech to find a solution to deal with problems from road traffic,and various countries in the world have begun to develop the smart car autopilot system.The most critical technology in the vehicle visual navigation subsystem is computer vision,whose tasks are road identification and tracking.Many scholars both here and abroad have conducted research on the visual navigation.Some scholars just focused on monocular vision technology and others researched into binocular or multi-dimensional visual navigation technology.In this thesis,first of all,the road images captured by three cameras on the experimental vehicle are preprocessed.According to the position and direction of the left,middle and right camera,ROI extraction is used to select the region of the image.Grayscale weighted method is employed to grayscale the ROI region image.And by using the histogram equalization,the effective information is prominent and the invalid information is eliminated,thus reducing the complexity of the image information;Gaussian filter is used to suppress the noise in the images,and to a certain extent reduce the noise interference to the image.Sobel algorithm is applied to detect the edges of the lane line in the images,which reduces the blurring degree of the image edges and improves the accuracy of the lane line edges to some extent.Secondly,this thesis is aimed to detect and identify the lanes and track the lanes.Therefore,in this thesis,the straight road model matching methods is adopted to match the road,and the curved road is simulated as the straight road for matching.Hough Transform Algorithm converts the rectangular coordinates of image space into the polar coordinates of parameter space,and tracks the lane lines of multi-frame images,adds a voting mechanism to detect the false lane lines,and sends the detected results to the upper module.Finally,a new lane reference line is fitted based on the three-eye four-lane line fusion algorithm.Lane line fusion is the process of judging the three-eye lane lines twice,that is,the first judgment is to determine whether the lane lines are found after the detection of Hough lane changing and it aims to eliminate non-lane interference lines.The second judgment is to determine whether they accord with the practical situation when the lane lines in the image are converted into the lane lines in the world coordinate and to prevent the loss of the lane detection because of the wrong camera placement or incorrect image acquisition.After the second judgment,the lane lines are matched with 16 possible reference lines,which are merged into the lane reference lines.