Lane Line Detection Based On Multiple Complex Road Scenes

In recent years,with the gradual development of the transportation industry,traffic accidents have also increased year by year,making traffic safety issues have received widespread attention from the society.In order to reduce the occurrence of traffic accidents and ensure driving safety,scholars have carried out a series of studies on car-assisted driving.Research shows that image-based lane line detection technology and lane departure warning technology have become the main research directions in vehicle assisted driving systems.The thesis mainly studies the problems of lane line detection in the vehicle assisted driving system,including the lane line is blocked by other obstacles;the trees and buildings on both sides of the road cause road shadows;the lane line fades,the color is inconsistent,and the lighting is uneven.Lane line detection algorithm in road scenes.For different lane line types and scenes,different algorithms are used for fitting.The main contents are as follows:1.The original image taken by the vehicle-mounted camera is preprocessed to enhance the lane characteristics and prepare for the subsequent lane line recognition.The preprocessing process includes binarization,noise reduction for the target area,extraction of the region of interest,and threshold segmentation according to the edge characteristics of the lane line to distinguish the target lane line from the background.2.For road scenes where the lane line type is a straight line and the illumination is normal,the Hough transform algorithm is used for fitting.First,the Canny operator based on double threshold gradient is used for edge detection,and then the Hough transform algorithm is used to fit the edge pixels of the lane line.The detection speed of this algorithm is fast and the real-time performance is good.3.For complex road scenes such as the type of lane line is curved,the road shadow is large,and the lane line fades,the sliding window polynomial algorithm is used to fit the lane line.The thesis improves the segmentation method combining multiple thresholds to separate the target from the background;finally,the curve is fitted using the sliding window polynomial fitting method.The results show that the algorithm can not only accurately identify the boundary and contour of the lane line in various complex scenarios,adapt to the curvature of the lane line,but also meet the real-time requirements.4.For multi-lane lines and road scenes where lane lines are blocked by other obstacles,the problem of multiple lane line detection is converted into an instance segmentation problem,and different lane lines are obtained through segmentation to form independent instances.First,use the Tu Simple data set to train a multi-task network Lane Net containing two branches;then use the H-Net network to achieve a bird’s-eye view of the image;finally use a polynomial to fit the multi-lane line.Compared with traditional image processing methods,this algorithm is more adaptable to complex scenes such as lane lines being occluded and unclear lanes ahead.It also improves robustness and has good detection accuracy.