Research On Lane Line Detection And Lane Departure Warning Technology Based On Monocular Vision

Intelligent driving is a general trend of the global automotive industry reform,and lane line detection and lane departure warning technology is the eyes of intelligent driving cars,is an indispensable basic technology.This technology can reduce many accidents caused by the non-autonomous departure of vehicles and has very important research significance and application value.In this paper,lane line detection and lane departure warning technology based on monocular vision was studied in-depth in order to further improve its detection speed and accuracy.The main research contents and results of this paper are as follows:(1)An efficient image preprocessing process(including gray-scale transformation,de-noising,enhancement,binarization and determination of detection region)is designed: according to the characteristics of road images,a color channel and weighted average method are designed to convert color images into gray-scale images;the gray mean was proposed as the image classification index and the images with different illumination intensity were enhanced;the detection area is determined according to road image features.The final detection area is about 17% of the original image,and the lane line features are more obvious.(2)The Hough lane line detection algorithm was improved: morphological filtering was used to remove the spots in the lane line;Canny custom difference operator was used for edge detection according to the characteristics of road image;according to the characteristics of lane line slope and width,the eligible lane lines are selected.In the absence of lane line tracking,the detection accuracy of this algorithm is about 87%,and the time is about 21 ms.(3)Propose a lane line detection algorithm based on dynamic partition detection region: according to the characteristics of lane line in the road image,the detection region was dynamically divided and the corresponding width threshold was set;according to the characteristics of lane line edge points(including gray value change,lane line width and distance between adjacent edge points,etc.),a multi-threshold lane line edge point extraction method was designed;RASANC based third-order Bessel curve was used to fit the extracted edge points.This algorithm can detect a variety of lane lines in a variety of complex environments very well,and its detection accuracy is about 5% higher than the improved Hough detection algorithm,and the time is about 11% less.(4)Design a lane line tracking algorithm based on conditional density propagation particle filter: firstly,the state transition model is determined according to the parameters of the lane line model;and then the model is updated;finally,the lane line model parameters are updated by resampling and state estimation,so as to realize lane line tracking.The tracking results are used to improve the lane line detection effect(detection accuracy is increased by more than 3% on the premise of ensuring real-time detection),thus improving the robustness of lane line recognition system.(5)The lane departure warning algorithm based on crossing lane line time was improved: design the calculation method of lane departure time;calibrate the on-board camera to obtain the internal and external parameters of the camera;compare the set time threshold to decide whether to give warning or not.The algorithm's deviation warning accuracy is about 94%,and the detection time of each frame is about 30 ms,so the algorithm has good practicability.