Research On Lane Line Detection Algorithms Of Autonomous Vehicles

The convenience and comfort brought by automatic driving is self-evident.More importantly,it can avoid many traffic accidents caused by human driving errors,and greatly protect the safety of people’s lives and property.Among them,lane detection as a basic and important part must meet the conditions that vehicles can obtain lane lines in real time and accurately in a variety of road conditions.Therefore,a lot of research work at present mainly focuses on its adaptability and robustness.In this paper,based on the principle of traditional image processing,under the condition of monocular vision,the algorithm for lane detection is studied.The classical Canny edge detection method can not automatically calculate the double threshold according to the characteristics of the current image edge adaptively is optimized and improved.Moreover,the method of feature information fusion is used to solve the problems of shadow and road brightness It’s the same problem.In the phase of lane detection,an adaptive sliding window method is proposed to filter Lane pixels,which improves the accuracy and accuracy of subsequent Lane fitting.This paper mainly studies the image preprocessing algorithm,edge detection algorithm and lane detection and tracking method.In the image preprocessing algorithm,the input image is processed by a series of operations to reduce the amount of subsequent calculation and remove the noise,including effective region extraction,graying and bilateral filtering.Finally,the filtered image is enhanced by histogram equalization to increase the contrast between lane line and image background,which is convenient for subsequent edge detection.In the phase of edge detection,the classical Canny edge detection method can only calculate the edge of all images by using the pre-determined fixed double threshold,so it can not be well adapted to the lane line edge detection in the changeable environment.Therefore,an adaptive dual threshold algorithm of Canny edge detection algorithm is designed.The edge direction in the image is estimated by the gradient vector of the image,and the dual threshold is calculated adaptively based on the gradient intensity of the estimated edge direction.At the same time,the input image is converted into HLS color space,and the lane information is extracted by using the feature that s channel is not affected by brightness.Finally,the results of the two methods are fused to filter out the shadow and uneven brightness in the road.For linear lane detection,this paper uses progressive probability Hough transform to recognize lane.Compared with the classical Hough transform method,it has better detection effect and algorithm performance.For the lane detection in the curve part,this paper first carries out the inverse perspective transformation on the input binary image,and then determines the starting point coordinates of the lane line according to the histogram of the transformed image.Then,the adaptive sliding window is used to filter out the lane line pixels.In the fitting part,in order to ensure the fitting accuracy and accuracy,the cubic curve model based on RANSAC algorithm is used The selected pixels are fitted.Finally,the Kalman filter algorithm is used to track the detected lane lines,and the correlation of video sequences is used to improve the recognition accuracy.