| With the development of science and technology,the automotive industry is striding towards the era of intelligent driving.Road identification and path tracking are the foundation of intelligent driving and the main component of Advanced Driver Assistance System(ADAS).This paper focuses on structured road identification.The aim is to propose a fast,accurate and robust lane line detection method.This paper has done the following research on the lane detection problem of intelligent vehicles:The vision-based lane detection technology mainly includes four aspects: image preprocessing,region of interest selection,lane line feature point extraction and lane line fitting and tracking.For the real road image collected by the sensor,this paper uses the grayscale transformation of the image to reduce the calculation amount of the image,and then uses the image enhancement technology to highlight the edge part of the image,and then uses the local OTSU algorithm to break up the lane line edge and noise in the image.The edge of the initial lane line is processed by the opening and closing operations to complete the preprocessing of the road image.The region of interest(ROI)selection is one of the important parts to improve the real-time and accuracy of road recognition.This paper uses affine transformation to transform the lane image into a aerial view.According to the road features in the aerial view,this paper proposes a dynamic region of interest(DROI)selection method based on horizontal and vertical safety distance,using the lane line equation and vehicle speed data of the previous frame to solve the curvature and maximum safe distance of each point on the road edge,through the transformation of the pixel and the actual distance,Realize the accurate division of the DROI at the current moment.According to the design characteristics of lane lines,this paper uses a lane line feature point recognition method based on DROI global search and improved Hough transform.Hough transform is better for straight road detection with small curvature,and DROI global search method is suitable for any road.In order to avoid the influence of far-distance noise in the image,the direction and range of the DROI Global Search are reasonably constrained.Then the chi-square distribution table method is used to eliminate the abnormal points in the feature point set,and the detection accuracy is improved.At the same time,considering the existence of the dotted line in the lane line type,this paper uses the primary and quadratic interpolation methods to interpolate the discontinuity of the line and the curve respectively.In order to achieve the fast and accurate matching of the lane feature points and the mathematical equations,this paper establishes the image polar coordinate system.The lane line is fitted by the polar coordinate equation and a good fitting effect is obtained.Finally,the classical lane line dataset Caltech is tested by this method,and the average positive detection rate is about 99.21%.Compared with the detection results of other mainstream methods at home and abroad,this method is more advantageous to others method.Further verification of the method,this paper further tests the real road image sequence with more complex conditions of 5683 frames,and finally obtains a good detection effect with a positive detection rate of 99.07%. |
