Research On Recognition And Tracking Method Of Lane Under Complex Road Conditions Based On Machine Vision

With the continuous development of science and technology,people’s living standards are constantly improving,and cars have become more and more important in our lives,which has brought us convenience and also led to a large number of traffic accidents.The carassisted driving system can effectively avoid traffic accidents,and the premise of carassisted driving is the correct identification and tracking of lane lines.In real life,there are complex and diverse driving environments,such as road congestion,missing lane lines,bad weather,and so on.It is not easy to accurately detect and track lane lines.Lane line extraction is mainly to detect and identify lane lines from images and determine the safe driving area of vehicles on the road.Whether the lane line is tracked correctly or not is related to the vehicle’s whereabouts at the next moment.Therefore,this paper proposes a method of lane detection and tracking in complex road conditions based on machine vision.The main research content is as follows:1.Aiming at the problem that the detection effect of straight lane lines is not good under complicated road conditions of structured roads,a lane line detection method based on cumulative probability Hough transform algorithm is given.A gradient calculation template is added in the direction of 45 and 135 of Canny operator,and the lane lines are constrained from two aspects: the lane line length and the angle between the lane line and the horizontal axis,and the lane lines are detected by combining with the Sobel straight line.Compared with the traditional Hough transform algorithm in different environments,the robustness and real-time performance of the algorithm are verified.2.In view of the poor effect of improved Hough transform on curve detection,an algorithm combining inverse perspective transform with sliding window curve detection is given.First,the road image is determined to be a curve,and the lane line that is larger than the set threshold is determined to be a curve.Secondly,the small curvature curve image is converted into a top view by inverse perspective transformation,which can approximate the curve in the top view as a straight line.Finally,the lane line is detected by combining the sliding window method and the parabolic model.3.Aiming at the problem that the environmental variables in the lane line detection image of unstructured roads are complicated and there are too many interference factors,according to the fact that the road area has the largest area,a regional growth algorithm based on the largest area is given to extract the road target area,and the non-target road area is eliminated by restricting the area,thus ensuring the accuracy of road boundary detection.Finally,the SIFT feature extraction algorithm is used to extract the feature points of the unstructured road boundary,and the extracted feature points are fitted by the least square method.4.For lane line tracking of structured roads,the traditional Kalman filtering algorithm is not ideal for lane line tracking,while unscented transformation can calculate the mean distribution and variance of the converted data through a few sampling points,thus calculating the covariance matrix,and then predicting the lane line position in the next frame.Therefore,based on Hough transform and unscented transform,a lane tracking algorithm based on unscented Kalman filter is proposed.It runs in various video sequences and compares with existing algorithms,which verifies the effectiveness of the algorithm on paper.