Interpretation of Lane Line detection Literature Series (i) A study on the detection algorithm of high-speed lane marking line based on machine vision _ Han

Author background

Research on the detection algorithm of high-speed lane mark based on machine Vision _ Han

    东北大学车辆工程硕士学位论文    2006年    7714】李晗. 基于机器视觉的高速车道标志线检测算法的研究[D2006. DOI:10.7666/d.y852642.`

Overview of the structure of papers

Grayscale of preprocessing

"Highlight" mode to determine whether to select day mode or night mode:

At the beginning of each detection cycle, first determine whether to use day mode or night mode work. The upper part of the camera's field of vision is the sky background, and the sky brightness can distinguish between daytime and nighttime environments. Because the sky is blue, the blue component of the daytime sky is the highest saturation, the algorithm only analyzes the blue component of the color image "extracts a fixed area of the sky (located in the middle of the image) and calculates the average luminance of the pixels within the area. If the value is below the specified threshold of 1, and the number of pixels in the whole frame image is higher than 200 (due to the high brightness of the rear lights at night) exceeds the threshold of 2, the night mode is used, otherwise the daytime mode is used.

using grayscale values that match human vision 0.30R,0.59G,0.11B
Enhanced image contrast with histogram equalization

In the case of dark light, the image is hard to discern what it looks like, after the histogram equalization many details can be easily seen clearly "but the histogram equalization increases the contrast and also increases the visual granularity of the image."
experiments on various methods of spatial and frequency domain filtering are carried out.

Binary (implemented by threshold segmentation)

• Histogram method (including P-one-digit method, Shuangfeng method, histogram concave analysis method)
• Maximum Inter-class variance (Ostu)
• Maximum Entropy method
• "Highlights" This paper uses the minimum error method to get the best threshold value
• "Highlights" two-valued road image correction: morphological filtering

Feature Extraction edge Detection

A comparison of several edge detection effects was done

• Roberts operator
• Sobel operator
• Prewitt operator
• Kriseh operator
• Guass a Laplace operator

Lane Extraction Inverse Perspective transformation

Lane recognition technology based on inverse perspective change image

First, the image is binary, and then based on the inverse perspective of the image reconstruction (remap remapping), so that the original perspective into a top view, the separation line from the intersection into a parallel, so that a certain algorithm to the top view of the separation line to search, and finally to the search for the separation line reverse transformation back to the original diagram, or directly to the top view as the basis for the detection of obstacles to the top view of the separation line detection algorithm has the use of ant agents (ant agent), template matching method. The difficulty of the algorithm is to calibrate the camera, find a suitable and effective inverse perspective transform function, and may lose useful information, the corner and slope can cause a large distortion of the reverse perspective transformation.

Regional Growth method

First, in the lower half of the image to detect the starting point of the line as the seed point of the region growth, and then according to their own growth criteria to grow "region growth of the seed point of determination, can be the historical data (that is, the previous time division of the separation line area) or empirical data, as well as the pixel itself gray value characteristics of the The determination of growth criterion is a difficult problem, because the gray value of the separation line in road image is not uniform. When the separation line is a virtual segment, it is also necessary to develop an algorithm to find the starting point of the next line.

Hough Transform

This algorithm usually first divides the image edge, then takes the edge point as the sample point and makes the Hough transform to find the separation line. The advantages of the algorithm are good anti-jamming, it is easy to detect the image of the linear target, but also can realize multi-lane detection "but the algorithm needs to open up additional Hough transform space, most researchers focus on the linear model of the detection research, for the curve detection needs to find a better separation line model, Hough Transform detection algorithm is not very ideal.

Template Detection Method
The detection and tracking of moving objects in the visual navigation of the document road vehicle

in this paper, the algorithm of Lane mark line extraction based on image restoration is adopted
Feature-based: Lane marking lines are rectangular areas of equal width
The specific algorithm does not look too clear.

Parameter fitting

Least squares

Interpretation of Lane Line detection Literature Series (i) A study on the detection algorithm of high-speed lane marking line based on machine vision _ Han