Lane Detection And Tracking For Complex Road Conditions Based On Filtering And State Estimation

With the transformation of traditional fuel vehicles to new energy vehicles,the functions of today’s vehicles are becoming more and more perfect,and unmanned driving,as an important research topic,is also booming.As one of the key technologies in unmanned driving,lane line detection technology is widely used in the fields of lane keeping,lane departure warning and path planning,and its algorithm types are also emerging in endlessly.However,the task of lane line detection in complex scenes is still arduous,and how to improve the robustness and accuracy of lane line detection results in complex road conditions is still a topic worthy of attention.This paper aims at lane line detection under complex road conditions.The main contents are lane line information extraction based on image preprocessing and multi-feature fusion,lane line fitting based on improved sliding window method and random sampling consensus algorithm,and lane line detection based on Kalman filter.Design the line tracking algorithm,and finally compare and verify the experimental results in the public dataset.1.Lane line information extraction based on image preprocessing and multi-feature fusion.The image preprocessing steps used in this paper include image undistortion and adaptive region of interest calculation to effectively remove distortion and noise in images under complex road conditions.At the same time,in order to extract complete and accurate lane line information,this paper designs a multi-feature fusion rule that combines different color information and edge information.2.Lane line fitting based on inverse perspective transformation and random sampling consensus algorithm.First,the image inverse perspective transformation technology is used to convert the image into a bird’s-eye view to calculate the coordinate parameters of the lane line in the real world.In this paper,considering the historical information of lane lines,an improved sliding window method is designed to locate the pixel coordinates of lane lines more accurately.Finally,based on the least squares method,this paper adds a random sampling consensus algorithm to eliminate the influence of interference points in the image,so that the lane line fitting results are more accurate.3.Design of lane tracking algorithm based on Kalman filter.In this paper,the Kalman filter algorithm and the extended Kalman filter algorithm are used to model the lane line model,considering the problems such as the loss of lane line information that is prone to occur in complex road conditions.Firstly,a multi-dimensional Kalman filter lane tracking algorithm for linear objects is designed to realize real-time robust tracking of the parameters of the lane line equation;then a lane line tracking algorithm based on extended Kalman filter is designed to realize the key coordinate points in the lane line Tracking and prediction make the overall lane tracking algorithm more robust.4.The public dataset is used to compare and verify the lane detection and tracking algorithm designed in this paper.The experimental results show that the lane line detection and tracking algorithm in this paper has an accuracy rate of 85.89% and a detection rate of 98.63% under complex road conditions,which has good robustness and high accuracy.