OpenCV 차선 인식 주요 기능 정리

Overview

We will explore OpenCV’s capabilities and examine which lane detection methods are used in various projects,
as well as determine the most suitable approach for implementing autonomous driving.

Key Features

• HSV Color/Saturation/Luminance Conversion

• Grayscale Conversion

• Canny Edge Detection

• Region of Interest (ROI) Definition

• Hough Transformation - Line Detection

• Gaussian Blur - Noise Reduction 

Let’s take a look at each of these key features one by one.

1. HSV Color/Saturation/Brightness Conversion

HSV is a method of color recognition based on hue, saturation, and value. You are likely familiar with the RGB method, which recognizes color using the three primary colors: Red, Green, and Blue. The reason we use HSV over RGB is, first, that it is less sensitive to changes in lighting. With the RGB method, all three values begin to fluctuate when lighting changes. However, since HSV separates brightness (V) and hue (H), it can produce more stable results even for the same color. In fact, when I tested this by writing code, I found that while the RGB method fails to recognize white when a shadow is cast over it, HSV recognizes it as white without issue. 

2. Grayscale Conversion

Grayscale conversion is a method available in the RGB system that converts values represented by the three RGB components into a grayscale image. This conversion is often used to speed up computational processing and thereby accelerate the detection of driving lines. In the HSV system, additional processing tends to slow things down, so we use only the basic HSV values.

3. Canny Edge Detection

Canny Edge Detection is a function that detects edges in an image. The principle behind finding these edges involves detecting areas where color changes abruptly, allowing only the image edges to be detected. Similarly, this method has the advantage of improving computational processing speed.

4. Region of Interest (ROI)

Region of Interest (ROI) designation is a function that selects only specific areas to include in the processing scope. In fact, processing speed is arguably the most critical factor in autonomous driving. Therefore, this feature allows you to designate a specific area—typically the lower center—as the ROI to detect lane markings or where the lane curves, offering the advantage of faster detection compared to processing the entire image.

5. Hough Transformation for Line Detection

The Hough Transformation is a function used to detect straight lines in an image. This function is used to verify whether lane markings are straight, identify surrounding structures, and determine the vehicle’s current direction.

6. Gaussian Blur for Noise Reduction

Gaussian Blur is used to remove noise that occurs in images due to environmental factors or camera specifications. Since the previously mentioned Kenny edge detection process recognizes every boundary as a separate point, this can slow down computations or cause errors; therefore, this feature is used to remove such noise.