CN111783666A - A Fast Lane Line Detection Method Based on Corner Feature Matching of Continuous Video Frames - Google Patents

Abstract

The invention relates to a fast lane line detection method based on corner point feature matching of continuous video frames, comprising: calculating an image distortion matrix; performing image acquisition and image correction; If the result of the lane line recognition in the previous frame is failed, select the region of interest as global and skip this step; find the coordinate values of the feature point pairs of the video frames before and after; if the lane line recognition result of the previous frame image is true, Perform optical flow estimation on the points with significant corner features in the previous frame image, and use the least squares solution to obtain the coordinate values of the feature point pairs of the video frames before and after; Area of interest; perform image binarization processing in the area of interest to obtain a binarized image; perform perspective transformation on the image; find lane line pixels in the top view, and use polynomial equations to fit lane lines; statistics lane curvature and lane The offset distance, annotated into the original image.

Description

A Fast Lane Line Detection Method Based on Corner Feature Matching of Continuous Video Frames

technical field

The invention belongs to the technology of environment perception and image processing in intelligent vehicles, in particular to a fast lane line detection method based on corner feature matching of continuous video frames.

Background technique

According to the statistics of the Ministry of Public Security, in 2019, there were 32.14 million newly registered motor vehicles nationwide, and the number of motor vehicles reached 348 million. The number of private cars exceeded 200 million for the first time, reaching 207 million. China's total highway mileage has reached 4,846,500 kilometers, and highways have reached 142,600 kilometers, ranking first in the world. The continuous growth of car ownership and total road mileage has increased the frequency of traffic accidents; at the same time, due to the relative concentration of vehicle travel areas, the degree of traffic congestion has been exacerbated. According to statistics from the Traffic Administration of the Ministry of Public Security, in 2019, a total of 238,351 road traffic accidents occurred nationwide, resulting in 67,759 deaths, 275,125 injuries and direct property losses of 910 million yuan. Frequent traffic accidents endanger people's life and property safety, and at the same time cause a waste of social resources and cause serious direct and indirect economic losses.

Therefore, countries have given strong support to the Intelligent Transportation System, and at the same time, companies and scientific research structures around the world have also invested a lot of manpower and material resources in research. Among them, intelligent vehicles (Intelligent Vehicles), as a key component of the intelligent transportation system, play a role that cannot be ignored. Intelligent vehicles perceive the environment around the vehicle in real time through sensors such as on-board cameras and radars, and resume local maps through algorithms based on the information obtained by the sensors, and then control vehicle actions through intelligent software systems to make vehicles safer and more reliable on the road.

Due to the limitations of the existing technology level and network transmission speed, the vehicle is not yet able to achieve fully autonomous driving. Many research institutions first realize the Advanced Driver Assistance System (ADAS, Advanced Driver Assistance System). ADAS integrates the perception of the surrounding environment of the vehicle and vehicle control, and realizes some basic functions of intelligent driving, including automatic cruise, automatic collision prevention, lane departure warning and automatic parking. Among the many sensors in the body, cameras are widely used because they are cheap and contain rich color information, and are mainly used to detect information such as lane lines, vehicles, road signs, and pedestrians around the vehicle.

The current lane line recognition algorithms mostly use the Hough operator method and the linear equation fitting method. These methods are practicable and can fit the position of the lane line on a straight road, but the Hough operator method is easy to lose the information of the curved road. The linear equation fitting method requires good edge features of the lane line as a premise, and because there is no global constraint, the susceptible to noise.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a fast lane line detection method based on continuous video frame corner feature matching, which is used to solve the problems of lack of curvature information and high complexity in the existing lane line identification method,

The present invention is a fast lane line detection method based on continuous video frame corner feature matching, which includes: step 1, calculating an image distortion matrix; step 2, performing image acquisition and image correction; step 3, using corner detection and matching The method of delineating the dynamic area of interest includes: if the result of the lane line recognition in the previous frame is failure, select the area of interest as the global one, and skip step 3; obtain the coordinate values of the feature points of the video frames before and after; When the result of identifying the lane line in one frame of image is true, the optical flow is estimated for the points with significant corner features in the previous frame of image, and the least squares solution is used to obtain the coordinate values of the feature points of the video frames before and after; The position of the lane line in the frame image is selected to identify the region of interest; step 4, image binarization processing is performed in the region of interest to obtain a binarized image; step 5, the image is subjected to perspective transformation to obtain a top view; step 6 , find the pixel points of the lane line in the top view, and use the polynomial equation to fit the lane line; step 7, count the curvature of the lane line and the lane departure distance, and mark them in the original image; step 8, repeat steps 2 to 6 until the image acquisition fails Or a termination identification signal is received.

The present invention proposes a fast lane line detection method based on feature matching of corner points of consecutive video frames. Corner points are intersections between contours. For the same scene, even if the viewing angle changes, it usually has the characteristics of stable quality. While retaining the important features of image graphics, corner points can effectively reduce the amount of information data, make the information content very high, effectively improve the speed of calculation, and be conducive to reliable image matching, making real-time processing possible.

Description of drawings

Fig. 1 is the method flow chart involved in the present invention;

Figure 2 is the image after distortion correction;

Figure 3 is an image after grayscale and binarization;

Fig. 4 is the image after perspective transformation;

Figure 5 is the image after lane line fitting;

Figure 6 is an annotated image.

Detailed ways

In order to make the purpose, content, and advantages of the present invention clearer, the specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

Fig. 1 is a flow chart of the method involved in the present invention; Fig. 2 is an image after distortion correction; Fig. 3 is an image after grayscale and binarization; Fig. 4 is an image after perspective transformation; Figure 6 is the marked image, as shown in Figure 1 to Figure 6, a fast lane line detection method based on continuous video frame corner feature matching of the present invention, including the following steps:

Step 1, calculate the image distortion matrix

Due to the deviation of the camera lens during the manufacturing and assembly process, distortion is introduced, resulting in distortion of the original image captured. Therefore, in the process of use, we need to correct the collected images by means of calibration.

Step 1.1, use the interface provided by the camera driver to capture images, and capture chessboard calibration images from 20 different angles

Step 1.2, use the weighted average method to grayscale the image

Step 1.3, using Zhang Zhengyou's calibration method, find the inner corner points of the calibration plate in the grayscale image and calculate the distortion matrix.

Step 2, Image Acquisition and Image Correction

At the beginning of each frame of image processing cycle, it is first necessary to determine whether the current processing cycle is the initial processing.

Call the interface provided by the camera driver to collect the image in front of the vehicle, and use the distortion matrix to perform distortion correction on the collected image to obtain the corrected image, as shown in Figure 2.

Step 3, use the method of corner detection and matching to delineate the dynamic region of interest (ROI, Region of Interest)

Step 3.1, if the result of lane line recognition in the previous frame is False, select the region of interest as the global one, and skip this step 3

Step 3.2, find the coordinate values of the feature points of the video frames before and after

When the result of identifying the lane line in the previous image is True, the L-K method is used to estimate the optical flow of the points with significant corner features in the previous image. Since the optical flow estimation equation is an overdetermined linear equation, the least squares method is used. Multiply the solution to obtain multiple sets of coordinate values of the feature point pairs of the video frames before and after.

Step 3.3, estimate the position of the lane line in the image of this frame, and select and identify the area of interest

Step 3.3.1, according to the position of the coordinates relative to the vertical line in the image, divide the feature point pair coordinate values into two groups: left and right.

Step 3.3.2, culling foreground pixels

For the left and right two sets of coordinate points, the DBSCAN clustering algorithm is used for the coordinate point displacement vectors respectively, and the foreground pixels are eliminated to obtain the main background coordinate points.

Step 3.3.3, use the weighted mean method to solve the relative displacement vector of the front background image

Step 3.3.4, set the region of interest

分别计算左右两侧车道的预估曲线，并将预估曲线的左右宽度为b(b＝50)的滑动窗口设定为感兴趣区域。Use the obtained displacement vector

Calculate the estimated curves of the left and right lanes respectively, and set the sliding window with the left and right width of the estimated curve as b (b=50) as the region of interest.

Step 4: Perform image binarization processing in the region of interest to obtain a binarized image

Step 4.1, use the composite operator to grayscale the image

Among them, the calculation method of Sobel operator is:

为平面卷积操作，A为原始图像。where Gx and Gy are the images of edge detection in the horizontal and vertical directions, respectively,

is a plane convolution operation, and A is the original image.

The formula for calculating the magnitude and gradient direction of the Sobel operator is:

where G and θ are the corresponding amplitude and gradient direction of the pixel, respectively.

Step 4.1.1, use the Sobel operator to grayscale the image in the horizontal direction, that is, obtain the image corresponding to Gx

Step 4.1.2, use the Sobel operator to grayscale the image on the amplitude value, that is, obtain the image corresponding to G

Step 4.1.3, use the Sobel operator to grayscale the image in the gradient direction, that is, obtain the image corresponding to θ

Step 4.1.4, take the image saturation channel component

Separate the hue (Hue), saturation (Saturation), and lightness (Lightness) channel components of the image in the HSL color space, and take the saturation channel component.

Step 4.2: Average the four grayscale images to obtain the grayscale image of the composite operator, as shown in Figure 3.

Step 5: Perform perspective transformation on the image to obtain a top view. As shown in Figure 4.

Step 6: Find the lane line pixel points in the top view, and use the polynomial equation to fit the lane line

Step 6.1, find lane line pixels

Step 6.1.1, calculate the histogram of the lower half of the image, and count the peak positions of the histogram on the left and right sides

Step 6.1.2, segment the image

The image was horizontally divided into 9 equal parts, and two rectangular sliding windows of equal height and width of 200 pixels were used in the bottom slice to cover the left and right peak positions of the histogram.

Step 6.1.3, find lane line pixels

Move the sliding window from the bottom to the top, find the lane line pixels in each slice in turn, and reposition the center of the sliding rectangle in the upper slice.

Step 6.2, second-order polynomial fitting using least squares

的公式为：For the left and right groups of lane line pixels, use the least squares method to perform second-order polynomial fitting respectively, and obtain the lane line equation under perspective transformation. Lane Line Equation

The formula is:

The calculation equations of the polynomial coefficients a ₀ , a ₁ , and a ₂ are:

Among them, x _i and y _i are the horizontal and vertical coordinates of the pixel points of the i-th lane line after the search in step 6.1.3.

The positions of the fitted lane lines in the image are shown in Figure 5.

Step 6.3, count the lane line pixel parameters, and mark the lane line recognition status bit

When the number of pixels is less than the set threshold of 200 or the curvature and range of the fitted curve exceeds the threshold range, the detection is considered to have failed, and the lane line recognition status bit is marked as False (by marking, the next frame will use global detection). This frame uses the detection result of the previous frame of video; otherwise, the lane line recognition status bit is marked as True.

Step 7: Count the curvature of the lane line and the distance of the lane departure, and mark the above information into the original image

Step 7.1, calculate the lane curvature and convert the units to meters according to the corresponding relationship.

Step 7.2, calculate the lane departure distance, and convert the unit to meters according to the corresponding relationship.

Step 7.3, information labeling.

On a blank image with the same height and width as the perspective transformed image, mark the left and right lane lines and green the middle area. The image is fused with the original image after inverse perspective transformation, and finally the curvature of the lane line and the distance of the lane departure are marked in the upper left part of the image with text. The annotated image is shown in Figure 6.

Step 8, repeat steps 2 to 6 until the image acquisition fails or a termination identification signal is received

The present invention provides another embodiment of a fast lane line detection method based on continuous video frame corner feature matching, comprising the following steps:

Step 1, calculate the image distortion matrix

Due to the deviation of the camera lens during the manufacturing and assembly process, distortion is introduced, resulting in distortion of the original image. Therefore, in the process of use, we need to correct the collected images.

Step 1.1, use the interface provided by the camera driver to take chessboard calibration images from 20 different angles

Step 1.2, use the weighted average method to grayscale the image

f(i,j)=0.3*R(i,j)+0.59*G(i,j)+0.11*B(i,j)

Among them, f(i,j) is the corresponding grayscale value of the pixel point whose coordinate is (i,j) after grayscale, R(i,j), G(i,j), B(i,j) respectively is the red, green, and blue channel components of the pixel with coordinates (i, j) in the color image.

Step 1.3, use Zhang Zhengyou's calibration method to find the inner corner points of the calibration plate in the grayscale image and calculate the distortion matrix

Step 2, Image Acquisition and Image Correction

At the beginning of each frame of image processing cycle, it is first necessary to determine whether the current processing cycle is the initial processing.

Call the interface provided by the camera driver to collect the image in front of the vehicle, and use the distortion matrix to perform distortion correction on the collected image to obtain the corrected image, as shown in Figure 2.

Step 3, use the method of corner detection and matching to delineate the dynamic region of interest (ROI, Region of Interest)

Step 3.1, if the result of lane line recognition in the previous frame is False, select the region of interest as global and skip this step

Step 3.2, perform optical flow estimation and find the least squares solution

When the result of identifying the lane line in the previous image is True, the L-K method is used to estimate the optical flow of the points with significant corner features in the previous image. Since the optical flow estimation equation is an overdetermined linear equation, the least squares method is used. The multiplication solution is:

where u and v are the relative displacement coordinates of feature points between video frames, and the calculation methods of each symbol are as follows:

After the calculation, multiple sets of coordinate values of feature point pairs of the video frames before and after are obtained.

Step 3.3, estimate the position of the lane line in the image of this frame, and select and identify the area of interest

Step 3.3.1, according to the position of the coordinates relative to the vertical line in the image, divide the feature point pair coordinate values into two groups: left and right.

Step 3.3.2, culling foreground pixels

For the left and right two sets of coordinate points, the DBSCAN clustering algorithm is used for the coordinate point displacement vectors respectively, and the foreground pixels are eliminated to obtain the main background coordinate points.

Step 3.3.3, solve the relative displacement vector of the front background image

Use the weighted mean method to solve the relative displacement vector of the front background image. The calculation formula is:

为求得的前方背景图像的相对位移矢量，是第i组征像素点的相对位移矢量，权值是第i个特征像素点与车道拟合曲线的距离倒数。in

is the relative displacement vector of the obtained front background image, which is the relative displacement vector of the i-th group of feature pixels, and the weight is the reciprocal of the distance between the i-th feature pixel and the lane fitting curve.

Step 3.3.4, set the region of interest

分别计算左右两侧车道的预估曲线，并将预估曲线的左右宽度为b(b＝50)的滑动窗口设定为感兴趣区域。Use the obtained displacement vector

Calculate the estimated curves of the left and right lanes respectively, and set the sliding window with the left and right width of the estimated curve as b (b=50) as the region of interest.

Step 4: Perform image binarization processing in the region of interest to obtain a binarized image

Step 4.1, use the composite operator to grayscale the image

Among them, the calculation method of Sobel operator is:

为平面卷积操作，A为原始图像。where Gx and Gy are the images of edge detection in the horizontal and vertical directions, respectively,

is a plane convolution operation, and A is the original image.

The formula for calculating the magnitude and gradient direction of the Sobel operator is:

where G and θ are the corresponding amplitude and gradient direction of the pixel, respectively.

Step 4.1.1, use the Sobel operator to grayscale the image in the horizontal direction, that is, obtain the image corresponding to Gx

Step 4.1.2, use the Sobel operator to grayscale the image on the amplitude value, that is, obtain the image corresponding to G

Step 4.1.3, use the Sobel operator to grayscale the image in the gradient direction, that is, obtain the image corresponding to θ

Step 4.1.4, separate image saturation channel components

Separate the hue (Hue), saturation (Saturation), and lightness (Lightness) channel components of the image in the HSL color space, and take the saturation channel component.

Step 4.2: Average the four grayscale images to obtain the grayscale image of the composite operator, as shown in Figure 3

Step 5: Perform perspective transformation on the image to obtain a top view. As shown in Figure 4

Step 6: Find the lane line pixel points in the top view, and use the polynomial equation to fit the lane line

Step 6.1, find lane line pixels

Step 6.1.1, calculate the histogram of the lower half of the image, and count the peak positions of the histogram on the left and right sides

Step 6.1.2, horizontally divide the image into 9 equal parts, and use two rectangular sliding windows of equal height and 200 pixels wide in the bottom slice to cover the left and right peak positions of the histogram

Step 6.1.3, move the sliding window from the bottom to the top, find the lane line pixels in each slice in turn, and reposition the center of the sliding rectangle in the upper slice

Step 6.2, perform second-order polynomial fitting

的公式为For the left and right two sets of lane line pixels, use the least squares method to perform second-order polynomial fitting, and obtain the lane line equation and lane line equation under perspective transformation.

The formula is

The calculation equations of the polynomial coefficients a ₀ , a ₁ , and a ₂ are:

Among them, x _i and y _i are the horizontal and vertical coordinates of the pixel points of the i-th lane line after the search in step 6.1.3.

The positions of the fitted lane lines in the image are shown in Figure 5.

Step 6.3, count the lane line pixel parameters, and mark the lane line recognition status bit

When the number of pixels is less than the set threshold of 200 or the curvature and range of the fitted curve exceeds the threshold range, the detection is considered to have failed, and the lane line recognition status bit is marked as False (by marking, the next frame will use global detection). This frame uses the detection result of the previous frame of video; otherwise, the lane line recognition status bit is marked as True.

Step 7: Count the curvature of the lane line and the distance of the lane departure, and mark the above information into the original image

Step 7.1, Calculate Lane Curvature

According to the fitting equation of the left and right lane lines, the mean curvature of the lane lines corresponding to each ordinate in the lower half of the image is calculated separately, and the unit is calculated according to the corresponding relationship of 3.7 meters/700 pixels in the horizontal direction and 30 meters/720 pixels in the vertical direction. Convert to meters and average the left and right mean curvatures.

Step 7.2, Calculate the Lane Departure Distance

According to the fitting equation of the left and right lane lines, count the difference between the center position of the left and right lane lines corresponding to each ordinate pixel in the lower half image and the horizontal center of the image, and convert the unit to 3.7 meters/700 pixels in the horizontal direction. Meter.

Step 7.3, Information Labeling

On a blank image with the same height and width as the perspective transformed image, mark the left and right lane lines and green the middle area. The image is fused with the original image after inverse perspective transformation, and finally the curvature of the lane line and the distance of the lane departure are marked in the upper left part of the image with text. The annotated image is shown in Figure 6.

Step 8, repeat steps 2 to 6 until the image acquisition fails or a termination identification signal is received

The fast lane line detection method based on continuous video frame corner feature matching proposed by the present invention is mainly used for lane line recognition applications in assisted driving and automatic driving. The method detects the relative movement of the background in front of the vehicle by introducing the continuous video frame corner matching method, and dynamically predicts the range of the lane line according to the temporal correlation between the video frames to reduce the region of interest (ROI, Region of intrest). Then, the image is combined grayscale and binarized. Finally, perform a perspective transformation on the image, find the lane line pixels and fit the lane lines using a second-order polynomial equation.

Aiming at the problems of lack of curvature information and high complexity in the existing lane line identification methods, the present invention proposes a fast lane line detection method based on the feature matching of corner points of continuous video frames. The corner points are the intersections between the contours. For the same scene , which is usually characterized by stable properties even if the viewing angle changes. While retaining the important features of image graphics, corner points can effectively reduce the amount of information data, make the information content very high, effectively improve the speed of calculation, be conducive to reliable image matching, and make real-time processing possible.

The above are only the preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, without departing from the technical principle of the present invention, several improvements and modifications can also be made. These improvements and modifications It should also be regarded as the protection scope of the present invention.

Claims (10)

1. A fast lane line detection method based on continuous video frame corner feature matching, comprising: Step 1, calculate the image distortion matrix; Step 2, image acquisition and image correction; Step 3, use the method of corner detection and matching to delineate the dynamic region of interest, including: If the result of lane line recognition in the previous frame is failed, select the region of interest as the global one, and skip step 3; Find the feature point pair coordinate values of the video frames before and after; In the case that the result of identifying the lane line in the previous frame of image is true, the optical flow is estimated for the points with significant corner features in the previous frame of image, and the least squares solution is used to obtain the coordinate value of the feature point pair of the video frames before and after; Estimate the position of the lane line in the image of this frame, and select and identify the area of interest; Step 4: Perform image binarization processing in the region of interest to obtain a binarized image; Step 5: Perform perspective transformation on the image to obtain a top view; Step 6: Find the lane line pixel points in the top view, and use the polynomial equation to fit the lane line; Step 7: Count the curvature of the lane line and the distance of the lane departure, and mark them in the original image; Step 8: Repeat steps 2 to 6 until the image acquisition fails or a termination identification signal is received. 2. The fast lane line detection method based on continuous video frame corner feature matching as claimed in claim 1, wherein in step 1, calculating the image distortion matrix comprises: Use the interface provided by the camera driver to capture images, and capture chessboard calibration images from 20 different angles; Use the weighted average method to grayscale the image; Using Zhang Zhengyou's calibration method, find the inner corners of the calibration plate in the grayscale image and calculate the distortion matrix. 3. the fast lane line detection method based on continuous video frame corner feature matching as claimed in claim 1, is characterized in that, step 2 carries out image acquisition and image correction and comprises: At the beginning of each frame of image processing cycle, determine whether the current processing cycle is the initial processing, and if it is the initial planning, mark the result of identifying the lane lines in the previous frame as failed; Call the interface provided by the camera driver to collect the image in front of the vehicle, and use the distortion matrix to perform distortion correction on the collected image to obtain the corrected image. 4. the fast lane line detection method based on continuous video frame corner feature matching as claimed in claim 1, is characterized in that, rejecting foreground pixel comprises: For the left and right two sets of coordinate points, the DBSCAN clustering algorithm is used for the coordinate point displacement vectors respectively, and the foreground pixels are eliminated to obtain the main background coordinate points. 5. the fast lane line detection method based on continuous video frame corner feature matching as claimed in claim 1, it is characterized in that, estimate the position of lane line in this frame image, select and identify the region of interest including: According to the position of the coordinates relative to the vertical line in the image, the feature points are divided into two groups, left and right; Eliminate the foreground pixels to get the main background coordinate points; Use the weighted mean method to solve the relative displacement vector of the front background image; Set a region of interest; Use the obtained displacement vector

Calculate the estimated curves of the left and right lanes respectively, set the left and right width of the estimated curve as b, and set the sliding window as the region of interest. 6. the fast lane line detection method based on continuous video frame corner feature matching as claimed in claim 1, is characterized in that, step 4 comprises: The calculation method of Sobel operator is:

where Gx and Gy are the images for edge detection in the horizontal and vertical directions, respectively,

is a plane convolution operation, A is the original image; The formula for calculating the magnitude and gradient direction of the Sobel operator is:

where G and θ are the corresponding amplitude and gradient direction of the pixel, respectively; Use the Sobel operator to grayscale the image in the horizontal direction to obtain the image corresponding to Gx; Use the Sobel operator to grayscale the image on the amplitude to obtain the image corresponding to G; Use the Sobel operator to grayscale the image in the gradient direction, that is, obtain the image corresponding to θ Take the image saturation channel component; Obtain a grayscale image of the composite operator. 7. the fast lane line detection method based on continuous video frame corner feature matching as claimed in claim 1, is characterized in that, step 6 specifically comprises: Find lane line pixels, including: Calculate the histogram of the lower half of the image, and count the peak positions of the histogram on the left and right sides; The image is horizontally divided into 9 equal parts, and two rectangular sliding windows of equal height and 200 pixels wide are used in the bottom slice to cover the left and right peak positions of the histogram; Move the sliding window from the bottom to the top, find the lane line pixels in each slice in turn, and reposition the center of the sliding rectangle in the upper slice; Second-order polynomial fitting using the least squares method; Count the parameters of the lane line pixel points, and mark the lane line recognition status bit. When the pixel points are less than the set threshold of 200 or the curvature and range of the fitted curve exceed the threshold range, the lane line recognition status bit will be marked as failed. frame using the detection result of the previous frame of video; otherwise, mark the lane line recognition status bit as successful. 8. the fast lane line detection method based on continuous video frame corner feature matching as claimed in claim 1, is characterized in that, step 7 specifically comprises: Calculate the lane curvature and convert the unit to meters according to the corresponding relationship; Calculate the lane departure distance and convert the unit to meters according to the corresponding relationship; On a blank image with the same height and width as the image after perspective transformation, mark the left and right lane lines and green the middle area, fuse the image with the original image after inverse perspective transformation, and mark the curvature of the lane line and the lane departure distance with text to on the image. 9. The fast lane line detection method based on continuous video frame corner feature matching as claimed in claim 7, characterized in that, using least squares method to perform second-order polynomial fitting, comprising: For the left and right two sets of lane line pixels, use the least squares method to perform second-order polynomial fitting, and obtain the lane line equation and lane line equation under perspective transformation.

The formula is:

The calculation equations for the polynomial coefficients a ₀ , a ₁ and a ₂ are:

Among them, x _i and y _i are the horizontal and vertical coordinates of the i-th group of lane line pixels after searching for the lane line pixels in each slice. 10. The fast lane line detection method based on continuous video frame corner feature matching as claimed in claim 1, wherein, Step 3 performs optical flow estimation, and finding the least squares solution includes:

where u and v are the relative displacement coordinates of feature points between video frames, and the calculation methods of each symbol are as follows:

After the calculation, multiple sets of feature point pair coordinate values of the video frames before and after are obtained.

Images