Road Obstacle And Lane Detection Based On Binocular Vision

With the development of automobile manufacturing and intelligent transportation fields in my country,the level of automobile intelligence is increasing year by year.Under the circumstances of civilianization of automobile prices and the standardization of road conditions,automobiles have become an important tool in people’s daily work and life.However,the large-scale use of automobiles has also produced a series of negative impacts and potential safety hazards,such as serious traffic jams during rush hours and frequent traffic accidents.How to prevent and reduce the occurrence of traffic accidents through the safety assisted driving system has become the core focus of increasing attention in the academic circles.Among them,obstacle distance detection and lane line detection based on binocular vision are heated issues in this field.This paper designs and develops a visual perception system that obtains information about the surrounding environment of the vehicle through sensors to assist the driver in completing reasonable control of the vehicle.The main research contents are as follows:(1)Camera calibration and epipolar correction.According to the principle of camera imaging,the four coordinate systems are studied through the correlation of the pixel coordinates of the image coordinate system and the three-dimensional space coordinates of the world coordinate system on the basis of the positional relationship between the coordinate systems;the advantages and disadvantages of different camera calibration methods are analyzed,and Zhang Zhengyou Method for Calibration is finally selected for calibration;combined with the principle of binocular camera calibration,according to the single target calibration result and the position transformation relationship between the two camera coordinate systems,the corrected image is obtained after epipolar correction.(2)Image stereo matching.Area matching and feature matching are commonly used stereo matching methods.According to the requirements of the ranging speed during the driving of the vehicle,this paper selects the feature matching method with faster matching speed.Comparing the matching effects of the three feature matching methods of SIFT,SURF and ORB,a stereo matching method on account of improved ORB-RANSAC is proposed;the method of combining epipolar constraint rooted in binocular position information and feature matching founded on Hamming distance is used to delete mismatched points: the K-dimensional tree(kd-tree)-based order consistency constraint method of neighbor points is used to screen out the initial interior point set,and the iterative pre-check method is utilized to improve the matching speed of RANSAC.(3)Binocular distance measurement.Through analyzing the principle of binocular distance measurement,finding out the corresponding projection points of the spatial point in the left and right images,and calculating the relative parallax,the actual spatial position coordinates of the spatial point is obtained after the two-target positioning results are combined;high-precision ranging results are achieved through the utilization of the quadric surface fitting method by making the coordinates of the points to be matched accurate to the sub-pixel level.(4)Lane line detection.The method of weighting the b component and gray image in the Lab color space is used to enhance the contrast of the yellow lane line;according to the characteristics of the lane line in the image where the line is continuous and the gradient direction is the same,the appropriate mask is selected to traverse the edge points,and the continuous contour area of rising and falling edges are obtained while the non-continuous interference edge is eliminated;after the collected contour edge curve is determined,different fitting algorithms are used to fit the lane line and curve,and the non-lane contour is removed according to the vanishing point principle and the position range of the lane line.