Research On Lane Line Detection Based On Vision

In recent years,traffic accidents caused by driver fatigue driving are not uncommon.In order to improve driver’s driving comfort and reduce traffic accidents,intelligent assisted driving technology has always been one of the hot research topics.Lane line detection technology is one of the key technologies in intelligent assisted driving technology,it is widely used in lane keeping and lane departure warning systems,which is of great significance for reducing traffic accidents and improving safe driving.The lane line detection based on machine vision is easily interfered by information such as glare,tree shadows,road text,road repair,and forked road markings.Aiming at this problem,this paper takes the lane segment extracted by Hough transform as the basic feature.Firstly,the region of interest is delineated,the interference of irrelevant background information on the detection algorithm is avoided,and the background edge noise is filtered out by the direction priority method.A lane line edge compensation method is designed for the discretized lane line edge information.The optimization of the edge characteristics of the lane line is realized,and a good lane line edge line segment is obtained.In order to better fit different types of lane lines,this paper constructs an indeterminate type of lane line model based on the parabolic model and a random sampling consensus algorithm.First,the extracted feature image is inversely transformed to eliminate the perspective effect and obtain a bird’s-eye view of the lane line.The DBSCAN clustering algorithm is used to detect the lane line,and the lane line detection area is delineated according to the local maximum value of the characteristic histogram.When fitting,the RANSAC algorithm is used to sample and define the data points in the region,extract the curve model parameters,and fit the parabola model to achieve the fitting of different shape lane lines.In order to test the real-time and robustness of the proposed algorithm,the experimental test analysis was carried out in the Caltech-lanes dataset and the Tusimple dataset.The experimental results show that the proposed algorithm has strong anti-interference ability and robustness,and the average processing time of each data frame of the two data sets is 65 ms and 69.5ms,which can basically realize real-time detection.