Lane Detection And Identifying Based On Monocular Vision

Statistics show that most of traffic accidents caused by drivers'drowsiness, inattentiveness. Driver assistance system is just used to monitor driver intent, warn drivers of lane departures, or assist in vehicle guidance, and avoid the traffic. The real-time lane detection and identifying system is also an essential module of the driver assistance and warning system. This paper is focus on detection and identifying of lane based on monocular vision.The thesis consists of video collection, image pre-processing, features extraction of lane, lane curve fitting and warning of lane departure. Based on Microsoft's VFW (Video for Windows) Video software and services package, this paper provides a new video interface, the realization of the real-time video streaming collection, acquisition static pictures ready to JPEG file format, which is the following processing algorithm based.Focused on detection of lane based on monocular vision, this paper adopts a new B-Spline 2D lane model, and assumes two sides of the road boundaries are parallel on the ground plane, this model can deal with more road constructions. The problems of detecting both sides of lane markings have been merged here as the problem of detecting the mid-line of the lane, by using the knowledge of the perspective parallel lines. As a result we make an effective improvement with the lane detection algorithm.Because the effect of edge detection of the road image is significant, this paper uses Canny operator to extract larger regional gradient changes. Due to the bend of the lane we split the lane area of interest into horizontal sections. This method takes advantage of perspective view's feature to find the horizontal (disappeared) line, and makes sure the lane area of interest.We make improvement on the basis of CHEVP(Canny/Hough Estimation of Vanishing Points)algorithm to extract the lane. We adopt an automatic adjusting method to determine the position of horizontal line and do the Hough transform. Thus we split the image correctly. We use gradient as the weight of the line, and make a constraint condition to removes the interferential lines, then get only one pair of lines in each image section as the candidate lane.We put forward a method to construct the mid-line according to the candid-ate lane, and adopt B-Spline model to fit the mid-line to a smooth curve .and do the same with both sides of the lane. At final we use the angle between the mid-line direction and the car's driving direction as the judgment of the lane departure warning.This method has been tested on video data and the experimental results have demonstrated a fast, robust system in real-time lane detection, and maybe do some help in the field of navigation based on vision.