A multi-lane line detection method based on equidistant edge point matching

Abstract

The invention discloses a multi-lane detection method based on equidistant edge point matching. The front road image is changed in reverse perspective projection to obtain a bird's-eye view of the road image, edge detection is performed on the bird's-eye view, and the detected edge map is preprocessed, and all edge points in the preprocessed edge map are localized Matching, keep the edge point pairs that match successfully, and obtain the feature point map through these edge point pairs, in the feature point map, divide the image into blocks according to the distribution of feature points, and extract all feature points containing a lane line, and Isolate the area belonging to other lane line feature points as a lane line candidate area. The invention can effectively solve the technical problems encountered in the detection of multi-lane lines, the positioning of curved lane lines, and the distinction of solid lane lines and virtual lane lines in the existing methods.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and more specifically relates to a multi-lane line detection method based on equidistant edge point matching. Background technique [0002] With the popularization of family cars, traffic safety issues have been paid more and more attention by people. According to statistics, in the world, the number of deaths caused by road traffic accidents is 500,000 people every year, which has become a major threat to life safety. [0003] In order to improve road safety and avoid traffic accidents, many research institutions and companies have conducted a lot of research on intelligent assisted driving, hoping to provide some auxiliary information before accidents to ensure traffic safety. The intelligent assisted driving system usually uses GPS, laser, radar, computer vision and other methods to cooperate with each other. Among them, the intelligent assistance method using computer vision has att...

AI Technical Summary

Problems solved by technology

However, most of the current lane line detection methods only detect the lane lines on both sides of the lane where the car is located. Although this reduces the difficulty of lane line detection, it loses a lot of important information.

[0007] 2. Unable to quickly and accurately locate curved lane lines

Among the existing methods, some use Hough transform to detect lane lines, which can only detect straight lane lines, and can only guarantee the detection results of a short distance ahead in the case of curved lane lines; some use parabolic models to detect curved lane lines. Lane fitting, although the parabolic model can fit simple turning situations and the calculation speed is relatively fast, but it is prone to deviation in complex road scenarios such as lane splitting and roundabouts; while using a more complex model, it is faced with The amount of calculation increases and the problem that cannot be detected in real time

[0008] 3. It is impossible to distinguish between real lane lines and virtual lane lines

In the actual driving process of a car, real lane lines and virtual lane lines have different meanings. Distinguishing them will help improve the results of lane line detection. However, due to the discontinuous nature of virtual lane lines, it is sometimes difficult to detect them. Detected, and there are few ways to distinguish them

Images