KR20140080105A - Method for detecting lane boundary by visual information - Google Patents

Abstract

The present invention relates a method for detecting a lane boundary by using image information. The method includes steps of inputting image information on a grayscale image; detecting a horizontal line by applying a vertical average distribution algorithm; detecting a lane boundary from an image at a lower part of the horizontal line; determining left and right lanes by applying a K-means clustering algorithm to the detected lane boundary, defining a lane derailment coefficient by setting a critical angle according to a change of a vanishing point, and determining a lane derailment or a lane change according to a vehicle′s driving state; and distinguishing an inlier from an outlier by applying an RANSAC algorithm and removing detection noise by repeatedly performing a lane detection function in consideration of only the inlier. Therefore, a lane change and a general drive may be distinguished from each other by using image information obtained from a moving vehicle, the reliability of a lane recognition result may be improved, and a lane may be detected accurately in harsh environments.

Description

TECHNICAL FIELD [0001] The present invention relates to a method of detecting a lane boundary using image information,

The present invention relates to a lane boundary detection method using image information, and more particularly, to a lane departure detection method using image information, which can distinguish a lane change from a normal lane using image information acquired from a vehicle while driving, And more particularly, to a lane boundary detection method using image information that enables accurate lane detection even in an environment.

Recently, there has been developed an intelligent vehicle in which a driving operation such as running, stopping, rotating, accelerating or decelerating is automatically performed by a computer or the like without directly operating the vehicle, even if there is no driver or there is a driver.

The main tasks of these intelligent vehicles are to maintain driving lanes, to establish safety distances to adjacent vehicles, to detect and avoid collision obstacles, to control collision avoidance, and to control vehicle speed in response to traffic conditions and road conditions.

Further, according to the progress of information and communication technology, a safety driving assist system such as a lane departure warning system (LDWS) or a lane keeping system, and a vehicle automatic control system have been developed, and the practical use of the intelligent vehicle is progressing more rapidly . Especially, the detection of the driving lane is one of the core technologies for solving the main problems in the intelligent vehicle, and many researches are actively conducted.

Since the detection of such a driving lane has a great influence on the safe driving, an accurate driving lane is detected by using various sensors in order to estimate and determine the position of the lane. For example, various sensors such as an image sensor, a RADAR or a LIDAR sensor have been used to implement an intelligent vehicle control system in a stand-alone or fused form for lane detection or object recognition in front of the vehicle.

Especially, image based system using image sensor has been widely used because it can extract a lot of information at low cost and can utilize various image processing algorithms.

Such an image-based lane detection system can extract feature information from an input image, apply a parametric model for lane detection, update algorithms such as matching, Kalman filter or particle filtering, A lane is detected using a method such as an approximation method or a non-parametric model matching by conversion such as a Hough Transform (HT).

As an example, Korean Patent Laid-Open Publication No. 10-2011-0001427 (2011.01.06) discloses a lane-line high-speed detection method by extracting a region of interest.

As shown in FIG. 1, the lane-based high-speed detection method using the ROI extraction method according to the related art includes a first step of pre-estimating a normal solid line from a focal length and an inclination angle parameter of a camera installed in a vehicle; A second step of optimally selecting a processing area necessary for lane detection from the vehicle to the normal solid line area and extracting ROI-LB (Region Of Interest for Lane Boundary) in a plurality of blocks; A third step of extracting feature information of an image in an area of interest and extracting a lane component from the image through a non-parameteric model matching technique based on Hough Transform (HT) based thereon; And a fourth step of removing noise included in the extracted lane components and detecting a lane through edge reinforcement.

Accordingly, when the lane is detected for the intelligent vehicle, it is possible to greatly reduce the lane detection speed and to improve the lane detection performance by reducing the image processing area through the estimation of the normal solid line and the selection of the area of interest A lane-line high-speed detection method by extracting a region of interest is provided.

However, in the lane-based high-speed detection method according to the related art as described above, it takes a long time to process the image during the normal solid line estimation and can recognize the lane in the general driving of the vehicle. However, There is a problem that reliability can not be distinguished from driving and therefore reliability can not be obtained and an accurate lane can not be detected in a severe driving environment.

Korean Patent Publication No. 10-2011-0001427 (2011.01.06)

SUMMARY OF THE INVENTION The present invention has been made to overcome the problems of the prior art described above, and it is an object of the present invention to provide a video processing apparatus and a video processing method, And RANSAC algorithm, it is possible to distinguish between lane change and normal driving by using image information acquired from a vehicle while driving, to improve the reliability of the lane recognition result, and to detect a lane more accurately even in harsh environments And to provide a lane boundary detection method using information.

According to an aspect of the present invention, there is provided a lane boundary detection method using image information, the method comprising: inputting image information of a gray scale image; A horizontal line detecting step of detecting a horizontal line by applying a vertical average distribution algorithm; A lane boundary detecting step of detecting a lane boundary from a lower end image of the horizontal line; A lane departure coefficient is defined by determining a left and a right lane by applying a K-means clustering algorithm to the detected lane boundary, setting a critical angle according to a change of the vanishing point, A lane change determining step of determining a lane change; And a RANSAC algorithm application step of discriminating between an inlier and an outlier by applying a RANSAC algorithm and a RANSAC algorithm to remove detection noise by repeatedly performing a lane detection function considering only the inlier, .

Wherein the horizontal line detecting step comprises: It is applied to the first frame of the lane boundary detection, and the position of the intersection between the two lane boundaries for the next frame can be detected as the horizontal line position.

The lane boundary detection step comprises: Selecting an image of the lower end of the horizontal line and converting it into black and white; Finding a white boundary point through a Canny edge image transformation; And extracting a straight line image of the boundary line and detecting a lane boundary.

Wherein the lane change determination step comprises: It is desirable to decompose a set of n objects based on distance into k clusters and determine the left and right lane boundaries of the screen through an integrated approach.

Wherein the lane change determination step comprises: Finds straight lines having the same direction as the lane, acquires information of the movement of the vanishing point at which the straight lines intersect, sets a critical angle using the k-means algorithm according to the obtained change of the vanishing point, The running state of the vehicle can be grasped.

Wherein the lane change determination step comprises: In the process of calculating the variable, it is preferable that the lane departure coefficient is defined and it is determined whether or not the vehicle deviates from the self-driving lane so as to grasp the running state of the vehicle.

According to the lane boundary detection method using the image information according to the present invention configured as described above, the image information acquired from the vehicle while driving is processed, the image processing time is shortened by detecting the horizontal line using the vertical average distribution algorithm, Using the -means clustering algorithm and the RANSAC algorithm, it is possible to distinguish between lane change and normal driving by using the image information obtained from the vehicle while driving, to improve the reliability of the lane recognition result and to detect the lane more accurately even in harsh environments There is an effect.

1 is a flowchart showing a lane-based high-speed detection method by ROI extraction according to the related art,
2 is a flowchart showing a lane boundary detection method using image information according to a preferred embodiment of the present invention.
3 is a flowchart showing the lane boundary detection step according to the present invention in more detail,
4 is a road image drawing for explaining a horizontal line detecting step during a daytime running according to the present invention,
5 is a road image drawing for explaining a horizontal line detecting step at night in accordance with the present invention,
6 is a road image drawing for explaining the lane boundary detection step of the present invention,
7 is a schematic diagram for explaining a K-means clustering algorithm,
8 is a flowchart for explaining a K-means clustering algorithm,
9 is a view showing a lane boundary and a vanishing point for explaining a lane change determination step of the present invention;
10 is an inlier and outlier distribution diagram for explaining the RANSAC algorithm of the present invention,
11 is a road image drawing for explaining lane detection during mist, rain, and night driving;
Fig. 12 is a view showing a plurality of road images during fog, rain, and night driving to show the accuracy of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. It is to be understood that one embodiment described herein is for illustrative purposes only and that the scope of the present invention is not limited to the description of one disclosed embodiment.

2 is a flowchart illustrating a lane boundary detection method using image information according to a preferred embodiment of the present invention.

As shown in FIG. 2, a lane boundary detection method using image information according to a preferred embodiment of the present invention includes an image information input step (S101) of inputting image information of a gray scale image; A horizontal line detection step (S102) of detecting a horizontal line by applying a vertical average distribution algorithm; A lane boundary detecting step (S103) of detecting a lane boundary from an image of a lower end of the horizontal line; A lane departure coefficient is defined by determining a left and a right lane by applying a K-means clustering algorithm to the detected lane boundary, setting a critical angle according to a change of the vanishing point, A lane change determining step (S104) for determining a lane change; And a RANSAC algorithm applying step (S105) for discriminating the inlier from the outlier by applying the RANSAC algorithm and repeating the lane detection function considering only the inlier to remove the detection noise.

Here, the horizontal line detection step (S102) includes: It is applied to the first frame of the lane boundary detection and detects the position of the intersection between the two lane boundaries for the next frame as the horizontal line position.

That is, the horizontal line is detected in the gray scale image. In the present invention, since the lane line is focused on the lower area of the horizontal line, the area above the horizontal line is not used. In particular, the minimum position is determined to be a horizontal line by utilizing a vertical average distribution method that determines the first minimum value that forms the upper curve.

The longitudinal average distribution equation can be expressed by Equation (1) below.

Here, VMD (i) is the average of the gray scale values of the ith row, H is the height of the original image, W is the width of the original image, Gr (i, j) And the intensity of the pixel in the jth column.

FIG. 4 is a road view for explaining a horizontal line detecting step during a daytime running according to the present invention, FIG. 5 is a road view for explaining a horizontal line detecting step at night in accordance with the present invention, Can be used to detect day and night horizontal lines.

The present invention detects a horizontal line through a vertical average distribution method as in Equation (1), and assumes a minimum position as a horizontal line by using a vertical average distribution method that determines an initial generated minimum value and an upper curve. Thus, the position of the intersection between the two lane boundaries for the next frame, which is applied to the first frame of the lane boundary detection, is regarded as the horizontal line position.

Therefore, the horizontal line detection step (S102) of the present invention is used to obtain only a substantially required lane, which is regarded as a sub-region image, so that the calculation time for the entire process is reduced and the upper portion of the image lane mark is removed So it is also effective for removing noise.

FIG. 3 is a flowchart illustrating a lane boundary detection step according to the present invention in more detail. As shown in FIG. 3, a lane boundary detection step (S103) according to a preferred embodiment of the present invention comprises: Selecting (201) an image of the lower end of the horizontal line and converting the image into black and white; A step (202) of searching for a white boundary point through the Canny edge image conversion; And a step (203) of detecting a lane boundary by extracting a straight line image among the boundary lines.

In FIG. 3, (a) shows a gray scale sub-regional image, (b) shows a canny image, (c) shows an improved edge image, and (d) shows k-means clustering.

Therefore, the lane boundary detection step (S103) of the present invention is used for searching for a boundary point of a lane and eliminating an unnecessary portion. By removing noise of components other than the lane before determining the left and right lanes, And accuracy can be increased.

FIG. 7 is a schematic diagram for explaining a K-means clustering algorithm, and FIG. 8 is a flowchart for explaining a K-means clustering algorithm.

In the lane change determination step (S104) of the present invention, a K-means clustering algorithm is applied as shown in the figure.

That is, the lane change determination step S104 may include: We decompose a set of n objects based on distance into k clusters and determine the left and right lane boundaries of the screen through an integrated approach.

In this case, the K-means clustering algorithm is a distance-based clustering method that divides a set of n objects into k clusters, and can be applied to determine a lane boundary through an integrated approach. Conventional approaches use a fixed width lane form, but the present invention detects left and right lanes, respectively.

7, a) k initial " mean "is selected in the random data set, b) the k clusters are created by concatenating all observations in the closest sense, and the partition representative Voronoi diagram C) setting the center k to the new means of each of the clusters, and d) repeating steps b) and c) until convergence is reached.

Here, the K-means clustering algorithm can be expressed by Equation (2) below.

Xi is the distance measurement selected from the data points, cluster center Cj is the distance index of N data points in the cluster center, and the number of clusters is K = 2. The two edge points are centered on the initial cluster of each lane Randomly selected.

The two objects can be classified through the algorithm as shown in FIG. 8 by using clusters of the distance values of the data point x and the cluster center point c.

At this time, in the lane change determination step S104, Finds straight lines having the same direction as the lane, acquires information of the movement of the vanishing point at which the straight lines intersect, sets a critical angle using the k-means algorithm according to the obtained change of the vanishing point, It is desirable to grasp the running state of the vehicle.

In addition, the lane change determination step (S104) may include: It is more preferable that the lane departure coefficient is defined in the process of calculating the variable to determine whether or not the vehicle has deviated from the own driving lane so as to grasp the traveling state of the vehicle.

That is, FIG. 9 shows a lane boundary and a vanishing point for explaining the lane change determination step of the present invention. In order to determine left and right lanes and define a lane departure coefficient, a K- Lane departure and lane change can be distinguished by determining the lane accurately and defining the lane departure coefficient by setting the critical angle according to the change of the vanishing point and identifying whether the lane change of the vehicle is the departure or lane change.

FIG. 10 is an inlier and outlier distribution diagram for explaining the RANSAC algorithm of the present invention. In the RANSAC algorithm application step (S105), the RANSAC algorithm is applied to distinguish between an inlier and an outlier, The detection function can be repeatedly performed to remove the detection noise.

At this time, 1) a minimum number of points required to determine the model parameters is randomly selected, 2) the parameters of the model are determined, and 3) a few points are selected that meet the predefined tolerance E. 4) If the fraction of the total number of points or more of the set or the number of the inliners exceeds the predefined threshold T, then apply all the inliners back to the model parameters and terminate; and 5) .

Therefore, by repeating the algorithm until the lane is recognized by obtaining the parameters of the lane, 1) Generally, when the data is taken in the natural phenomenon, noise is included a lot. In the case of the least squares method, The RANSAC algorithm according to the present invention has an advantage of being able to estimate a function considering only the inliers by separating the inliers and outliers, and 2) the input to the RANSAC algorithm is Reliability can be represented by model parameters that can be fitted or explained with observed data values and observations; and 3) each step of the RANSAC algorithm is repeated until the lane is detected, allowing accurate lane detection in difficult road environments.

Fig. 11 is a road image for explaining lane detection during fog (a), rainy day (b), and night (c), Fig. 12 is a view showing the fog (a) (C) at night, and Table 1 shows the results of the proposed algorithm.

Clip Total frame Detected frame Correct rate Frame / sec a 1500 1360 90.66% 27.0 b 486 458 94.24% 27.0 c 885 756 85.42% 28.0 Total 2871 2374 90.11% 27.3

Therefore, according to the lane boundary detection method using image information according to the present invention, image information acquired from a vehicle while driving is processed, and a horizontal line is detected using a vertical average distribution algorithm to shorten the image processing time , Using K-means clustering algorithm and RANSAC algorithm, it is possible to distinguish between lane change and normal driving by using image information acquired from a vehicle while driving, improve reliability of lane recognition result, and detect lane more accurately even in harsh environment can do.

The embodiments of the present invention described in the present specification and the configurations shown in the drawings relate to the most preferred embodiments of the present invention and are not intended to encompass all of the technical ideas of the present invention so that various equivalents It should be understood that water and variations may be present. Therefore, it is to be understood that the present invention is not limited to the above-described embodiments, and that various modifications may be made without departing from the spirit and scope of the invention as defined in the appended claims. , Such changes shall be within the scope of the claims set forth in the claims.

Claims (6)

An image information input step of inputting image information of a gray scale image;
A horizontal line detecting step of detecting a horizontal line by applying a vertical average distribution algorithm;
A lane boundary detecting step of detecting a lane boundary from a lower end image of the horizontal line;
A lane departure coefficient is defined by determining a left and a right lane by applying a K-means clustering algorithm to the detected lane boundary, setting a critical angle according to a change of the vanishing point, A lane change determining step of determining a lane change; And
A RANSAC algorithm applying step of discriminating between an inlier and an outlier by applying a RANSAC (Random Absolute Consensus) algorithm and repeating a lane detection function in consideration of only the inlier to remove detection noise; Lane boundary detection method using image information. The method according to claim 1,
Wherein the horizontal line detecting step comprises:
And detecting a position of an intersection between two lane boundaries for a next frame as a horizontal line position. The method according to claim 1,
The lane boundary detection step comprises:
Selecting an image of the lower end of the horizontal line and converting it into black and white;
Finding a white boundary point through a Canny edge image transformation; And
And detecting a lane boundary by extracting a straight line image among the boundary lines. The method according to claim 1,
Wherein the lane change determination step comprises:
And dividing the set of n objects based on the distance into k clusters and determining the left and right lane boundaries of the screen through an integrated approach. 5. The method of claim 4,
Wherein the lane change determination step comprises:
Finds straight lines having the same direction as the lane, acquires information of the movement of the vanishing point at which the straight lines intersect, sets a critical angle using the k-means algorithm according to the obtained change of the vanishing point, And detecting a traveling state of the vehicle. 6. The method of claim 5,
Wherein the lane change determination step comprises:
Wherein a lane departure coefficient is defined in the process of calculating the variable to determine whether the vehicle has departed from the self-driving lane, and to determine the running state of the vehicle.

Images