KR101347886B1 - Method and Apparatus for Road Lane Recognition by Surface Region and Geometry Information - Google Patents

Abstract

The present invention relates to a method and apparatus for recognizing a lane using road area and geometric information. The present invention provides a surface feature standard vector for a region of interest from a plurality of road images stored in a database, and compares the road region with a surface feature vector of an input image. Determining the lane boundary and lane using geometric information of the lane within the road area, calculating the position, angle, and moving speed of the lane boundary to determine the lane change direction, lane change start point, and lane change completion time. And calculating an estimated lane change time for calculating the estimated time for the same.

Description

Method and Apparatus for Recognizing Road Using Road Area and Geometric Information {Method and Apparatus for Road Lane Recognition by Surface Region and Geometry Information}

The present invention relates to a method and apparatus for recognizing a lane, and more particularly, to a lane recognition method and apparatus for recognizing lanes and lanes in a road area by using road area and geometric information from an image input by a driving vehicle.

  The IVS (Intelligent Vehicle System) field, in which automobiles become an indispensable means of modern life and adopts advanced IT technology, has become the most noticeable research field. Recently, among the IVS fields, a support system for driving safely around a vehicle by a sensor installed in the vehicle has been attracting attention. In particular, it is mainstream to analyze the situation occurring in front of the driving vehicle, but in the case of an actual accident, more accidents occur at the side and rear than the front of the vehicle.

Car crash accident analysis and statistical results are mainly dominated by drowsy driving or careless driving. However, with the exception of single accidents, the next most likely crash is a crash. Analysis of car crashes in the United States shows that about 8% of all highway crashes occur during lane changes. The biggest cause of such accidents is the driver's situational awareness and error. In order to reduce accidents, when changing lanes, the driver must be aware of the surrounding lane as well as the driving lane. In the case of a lane change, the basic position is the relative position information of the lane in which the driver is currently driving and the lane to be changed. In addition, based on the recognized lanes, situations such as vehicle presence and speed in each lane should also be recognized.

Conventionally, lane characteristics are defined using color information and detected edge angles, lane boundaries are detected using a probability model, and lane boundaries are recognized using geometric features such as vanishing points by perspective. However, such a prior art has a problem that it is impossible to accurately detect a lane marking because it does not take into account the characteristics of each lane itself.

Japanese Patent Laid-Open No. 5-151341 (1993.06.18) discloses a vehicle traveling path detecting device. FIG. 15 is a block diagram illustrating a vehicle traveling road detection apparatus according to the related art disclosed in the above publication. As shown in the drawing, a known vehicle traveling path detecting apparatus receives a signal from the image input means 100 for capturing an image of a road in front of a driving vehicle and the image input means 100 to extract a boundary point. Setting means 101 and setting the window near the linear equation obtained in the previous calculation operation and measuring the coordinates of the plurality of boundary points in the set window to obtain the multiple linear equation through linear calculation. The slope of the straight line through the vanishing point based on the line adaptation means 102, the X- and Y-coordinates of the vanishing point, and the linear equation obtained to minimize the sum of error squares in the linear equation. The line and vanishing point determining means 104, the result and line of the preparation means 101, which estimate the quantity corresponding to and also set these estimated values to the result at this time. An edge point tracing mean 103 for extracting a boundary point corresponding to the lane display based on the detection result of the response means 102, and a curve adaptation means for applying a curved equation to the extracted boundary point ( curve adaptation means 105 and smoothing means 106 for smoothing the result of the curve adaptation means 105.

Since the application of the curve equation is performed by the least square method, the known vehicle traveling road detection apparatus has a problem in that it is impossible to accurately detect lanes and lanes from an input image because it is easily affected by noise.

Japanese Patent Laid-Open No. 5-151341 (1993.06.18)

SUMMARY OF THE INVENTION The present invention has been made to overcome the above-mentioned problems of the related art, and an object thereof is to provide a method and apparatus for recognizing a lane and a lane from an input image.

Another object of the present invention is to provide an efficient lane and lane recognition method and apparatus by effectively selecting only information necessary for recognizing lanes and lanes from an input image.

The present invention also provides an accurate lane and lane recognition method and apparatus without being easily affected by noise by minimizing the effects of artifacts such as building exterior walls, soundproof walls, etc., or natural objects such as the sky or trees present in the input image. Its purpose is to.

In order to achieve the above object, the present invention provides a road region determination step of obtaining a surface feature standard vector of a region of interest from a plurality of road images stored in a DB, and comparing the surface feature vector of an input image with a road feature; A boundary line and a lane detection step of detecting an intersection with a lane using geometric information of a lane in a road area, acquiring a vanishing point, and recognizing a lane boundary and a lane; A lane area estimation time calculation step of determining a lane change direction, a lane change start time, and a lane change completion time by calculating a position, an angle, and a moving speed of a lane boundary, and calculating an estimated time for the lane change direction; Provides a method of recognizing a car using information.

The determining of the road area may include setting a region of interest according to the same distance in any one specific image in a DB including a plurality of road images; Dividing the region of interest into a plurality of partitions, and obtaining respective surface feature vectors for each partition; Calculating a mean for each surface feature vector to obtain a surface feature standard vector for the region of interest; Calculating a surface feature vector of the divided region in the input image and comparing the region with the surface feature standard vector included in the region of interest according to a corresponding distance to determine the road region.

The boundary line and lane detection step may include: obtaining an adaptive threshold value from a road area; Binarizing the image; Performing labeling using a pixel count and storing a blob of interest; Detecting a lane using lane individual features (spindle information); Acquiring the intersection using the main axis information of the blob of interest; Determining the highest density small region through intersection analysis; Obtaining a vanishing point based on the center of gravity of the small region; Establishing a lane boundary candidate through vanishing line acquisition and angle integration; And detecting a lane boundary line according to a position after the vanishing point and the lane, and recognizing the lane through the boundary line.

The lane change estimated time calculating step may include detecting a vehicle boundary line from an input image; Determining a lane change direction by calculating an angle of a vehicle boundary line overlapping the lane boundary line; Detecting an overlapping position and a degree of the vehicle boundary line and the lane boundary line and determining a lane change start time when the vehicle boundary line overlaps with a reference value or more; Predicting a lane change completion time by detecting a size and a moving speed of the vehicle; And calculating and displaying the estimated lane change time.

In addition, the present invention, the road region determination means for obtaining a surface feature standard vector for the region of interest from the plurality of road images stored in the DB, and determines the road area compared with the surface feature vector of the input image; Boundary line and lane detection means for detecting an intersection with a lane using geometric information of the lane within a road area, acquiring a vanishing point, and recognizing a lane boundary and a lane; Road area and geometry including; lane change estimated time calculating means for calculating the position, angle, and moving speed of a lane boundary, determining a lane change direction, a lane change start time, and a lane change completion time, and calculating an estimated time for the lane change direction Provided is a vehicle recognition apparatus using information.

According to the lane recognition method and apparatus using the road area and the geometric information according to the present invention configured as described above, the road area from the input image using any one or more information selected from the energy, contrast, homogeneity and correlation of the input image By detecting it, there exists an effect which can recognize a lane and a lane efficiently.

In addition, by accurately detecting lane boundaries and lanes using geometric information of lanes within the detected road area, the lane recognition rate is not easily affected, and the lane recognition rate is improved, so that the lanes can be detected more accurately.

1 is a control flowchart illustrating a lane recognition method using road area and geometric information according to an exemplary embodiment of the present invention.
2 is a control flowchart illustrating the road area determination step of FIG. 1 in more detail.
3 is a control flow diagram illustrating in detail the boundary line and lane detection step of FIG.
4 is a control flowchart illustrating the lane change estimated time calculation step of FIG. 1 in more detail.
5A and 5B are diagrams in which respective regions are divided, and each region of interest according to distance is represented by 256 colors.
6A-6D are graphs showing surface feature standard vectors of a region of interest according to distance for each feature.
7A and 7B illustrate an input image and one road candidate region determined therewith.
7C illustrates a road area determined as a result of an embodiment of the present invention.
8 is a flow chart illustrating an overview of an algorithm for detecting lane boundary lines from a road area according to a preferred embodiment of the present invention.
9A shows a binarization image, and FIG. 9B shows a labeling result.
10 is a graph illustrating a histogram and parameters for obtaining a threshold.
11 is a graph representing the obtained intersection point X _u (x, y).
FIG. 12 is a graph showing the intersection (highest density region in FIG. 10) within a circle of radius 20. FIG.
FIG. 13A is a vanishing line, and FIG. 13B is a lane boundary line detection result. FIG.
FIG. 13C is a diagram illustrating a lane currently being driven and a lane to be entered when the lane is changed.
14 is a block diagram illustrating an embodiment of a lane recognition apparatus using road area and geometric information according to the present invention;
15 is a block diagram of a vehicle traveling road detecting apparatus according to the related art.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a control flowchart illustrating a lane recognition method using road area and geometric information according to an exemplary embodiment of the present invention.

As shown, the lane detection method using the road area and the geometric information according to an embodiment of the present invention, to obtain the surface feature standard vector for the ROI from a plurality of road images stored in the DB, the surface feature of the input image A road area determination step (S010) of determining a road area by comparing with a vector; A boundary line and a lane detection step of detecting a crossing point with a lane using geometric information of a lane in a road area, acquiring a vanishing point, and recognizing a lane boundary and a lane (S020); A lane change estimation time calculating step (S030) is performed to calculate the position, angle, and moving speed of the lane boundary, determine a lane change direction, a lane change start time, and a lane change completion time, and calculate an estimated time for the lane change direction.

2 is a control flowchart illustrating the road area determination step of FIG. 1 in more detail. As shown, the step of determining the road area according to an embodiment of the present invention (S010), the step of setting the ROI according to the same distance in any one specific image in a DB including a plurality of road images ( S011); Dividing the region of interest into a plurality of divided regions, and obtaining respective surface feature vectors for each divided region (S012); Calculating a mean for each surface feature vector to obtain a surface feature standard vector for the region of interest (S013); Computing a surface feature vector of the region of interest in the input image and comparing the surface feature standard vector to determine a road region corresponding to the region of interest (S014). In addition, the target for obtaining the surface feature vector of the road area determination step (S010) according to the preferred embodiment of the present invention is preferably at least one data selected from image energy, contrast, homogeneity and correlation.

In a preferred embodiment of the present invention, the proposed algorithm for finding road areas obtains a surface feature vector of a segmented area in an image from a database composed of a plurality of road images (hereinafter, abbreviated as DB), The average of the surface feature vectors of all the divided regions included in the region of interest according to the distance defined accordingly is calculated. That is, preferably, the average of each surface feature vector defined for each region of interest according to distance may be a surface feature standard vector, which is a comparison reference value for determining a road region for each region.

In the input image, the road region is determined by calculating the surface feature vector of the divided region and comparing the surface feature standard vector included in the region of interest according to the corresponding distance with the region. The result of the road area thus obtained is used in the image binarization process to determine the area of interest color in the process of finding the lane later.

According to a preferred embodiment of the present invention, the ROI may be set according to the distance within the image in the database. The surface feature standard vector representing this region is determined by calculating the average for each surface feature vector in all divided regions distributed within the region of interest over the same distance.

Next, the segmentation area is determined in the image of the DB and the surface feature vector is calculated within this area. In this case, the divided region according to the exemplary embodiment of the present invention refers to each patch divided into 40 × 40 images 640 × 80. In the embodiments to which the present invention is applicable, the preferred size of the partition may be appropriately selected by those skilled in the art to which the present invention pertains. The surface feature vector according to an exemplary embodiment of the present invention may define a value for each region by dividing the region based on the distance in the image. At this time, the region divided according to the distance is defined as the region of interest according to the distance. This is defined by Equation 1 below and can be divided into 18 zones.

상기 수학식 1에서 Tr는 비교 영역을 결정하는 임계값, dv는 교차점 X_u 와 T_r 반경 내에 포함되는 모든 교차점과의 거리를 나타낸다.

In Equation 1, Tr represents a threshold for determining a comparison area, and dv represents a distance between all intersections included in the intersection X _u and the radius of T _r .

5A and 5B illustrate an example in which each region is divided according to a preferred embodiment of the present invention, and each region of interest according to a distance is represented by 256 colors.

In addition, in a preferred embodiment of the present invention, a feature vector for any one or more selected from energy, contrast, homogeneity and correlation may be used as a value representing the surface characteristic of each patch for the road area. Each surface feature is defined by Equations 1 to 5 below.

,

는 상생행렬(co-occurance matrix, 또는 공생행렬)에서 (i, j)의 변수(파라미터, 또는 인덱스)에 따른 각각의 표준편차를 나타낸다.
여기서 p(i,j)는 분할 영역에서 0도 방향으로 이웃한 2 픽셀간의 인텐시티(intensity) 변화를 나타내는 상생행렬(Cooccurrence matrix)에서의 동일 변화 빈도수를 나타낸다. 분할 영역 내 각 영역에서 4가지 표면 특징을 이용하여 도로 표면을 정의한다. 여기서 E, Ct, H, Cr은 각각 에너지(Energy), 콘트라스트(대비, Contrast), 균질성(Homogeneity), 상관도(Correlation)을 나타낸다.

,

Denotes each standard deviation according to the variable (parameter, or index) of (i, j) in the co-occurance matrix.
Here, p (i, j) represents the same frequency of change in the cooccurrence matrix representing the change in intensity between two pixels adjacent to each other in the direction of 0 degrees in the divided region. The road surface is defined using four surface features in each area within the partition. E, Ct, H, and Cr represent Energy, Contrast, Homogeneity, and Correlation, respectively.

In a preferred embodiment of the present invention, the surface feature standard vector, which is a reference value for detecting the road area, can be determined by calculating the average of the surface feature vectors located in the region of interest according to the same distance of all databases.

6A-6D are graphs showing the surface feature standard vectors of the region of interest according to distance for each feature, where FIG. 6A is energy, FIG. 6B is contrast, FIG. 6C is homogeneity, and FIG. 6D is a correlation diagram. If the index (horizontal axis) is 0 based on (0,0) of FIG. 5A described above, the index is closer to the origin of FIG. 5A, and the larger the number, the farther from the origin.

Next, the above-described region segmentation and surface feature vector detection processes are repeated on the input image to calculate the surface feature vectors of the divided regions of the input image.

In a preferred embodiment of the present invention, two out of four features correspond to ± 50% by comparing the calculated surface feature vector for each divided region with the surface feature standard vector at a position corresponding to the region of interest over distance. In this case, the divided region of the input image may be determined as the road candidate region.

7A and 7B illustrate an input image and one road candidate region determined with respect to the input image. In a preferred embodiment of the present invention, a morphology (close) operation may be performed on the result of FIG. 7B to determine one road area among the road candidate areas detected above. The result of the labeling process is to obtain independently separated regions, and to determine a single region having the largest number of pixels among the regions as a road region. 7C shows the road area determined through this process.

Labeling refers to an area classification operation performed on a binary image by attaching the same number to all adjacent pixels and different numbers to other unconnected components. The labeling is a general academic term commonly used in the field of image processing, and each term may be interpreted in a dictionary meaning, and thus a detailed description thereof will be omitted.

The boundary line and lane detection step (S020) according to an embodiment of the present invention; The intersection with the lane may be detected using geometric information of the lane within the road area, and the vanishing point may be obtained to recognize the lane boundary and the lane.

3 is a control flowchart illustrating in detail the boundary line and lane detection step of FIG. 1. As shown, the boundary line and lane detection step (S020) according to a preferred embodiment of the present invention; Obtaining an adaptive threshold value from the road area (S021); Binarizing the image of the road area (S022); Performing labeling using a pixel coefficient and storing a blob of interest (S023); Detecting a lane by using lane individual features (spindle information) (S024); Acquiring an intersection point using the main axis information of the blob of interest (S025); Determining the highest density small region through the intersection analysis (S026); Acquiring a vanishing point with the center of gravity of the small region (S027); Setting a lane boundary candidate through evanescent line acquisition and angular integration (S028); And detecting a lane boundary line according to a position following the vanishing point and the lane, and recognizing the lane through the boundary line (S029).

In the above description, a blob writes a binary type in a database, and is distinguished from a text (TEXT) for writing a character string in a database and a clob (CLOB) for writing a character type in a database. Binary data types can be files of any type. In general, text is used to store string data, and blob is a type for storing documents, images, and video clips. Hence, Blob has the advantage of being able to record all files that the computer recognizes, from normal word documents to media files. The blob is a general term commonly used in the field of image processing, and each term may be interpreted in a dictionary meaning, and thus a detailed description thereof will be omitted.

8 is a flowchart illustrating an outline of an algorithm for detecting lane boundary lines from a road area according to a preferred embodiment of the present invention.

In a preferred embodiment of the present invention, lane boundary lines are detected prior to lane recognition. For this purpose, lanes in the same direction are detected on the premise that lanes are drawn parallel to each other on the road plane. The lanes on the pavement are mainly white, which clearly distinguishes the asphalt area from the color. This color feature is important as a feature that can distinguish and detect only an area of a lane on a road. In a preferred embodiment of the present invention, a method of adaptively determining a threshold value may be used to distinguish a bright color region close to white and a remaining road region in an input image.

In a preferred embodiment of the present invention, the rectangular shape of which one side is long in the lane can also be utilized as a good feature for detecting the lane. To do this, blobs of independent regions are obtained through a labeling process in the binarized image. Principal axes are obtained by obtaining a covariance matrix to represent the distribution of pixels included in each blob and calculating their eigenvalues and eigenvectors. The main axis is used to define the lane shape. Parallel lanes meet at a single vanishing point according to perspective projection. The process of finding this vanishing point is described later. Following the extension lines of the major major axes of interest blobs satisfying the condition of color and shape, find the intersection point where the major axes meet and analyze them to find one vanishing point. A lane boundary is created by connecting the acquired vanishing point and the center of the blob of interest. The lane, which is the final result of the proposed algorithm, is recognized by considering the position of the camera installed in the vehicle based on the obtained lane boundary.

The lanes on the road are close to bright white, unlike the dark gray of asphalt. The input image is binarized to extract only the bright color region that is the characteristic of the lane. If the image contains a lot of unnecessary color areas such as natural objects such as sky or trees and artifacts such as buildings and soundproof walls, the appropriate threshold is not found in the process of finding a threshold for binarization. Therefore, in this experiment, the threshold value is determined in the road area obtained earlier to minimize the influence of unnecessary image information.

The threshold T is determined by Equation 6 and is a threshold that separates the background from the foreground in the input image histogram.

N ₁ and N ₂ are the total number of pixels in the histogram group divided by k, and H (i) is the histogram of the grayscale input image. And μ ₁ and μ ₂ are the mean of each group. By T, the binarized image I _b (x, y) is obtained from equation (7).

delete

9A is a diagram illustrating a binarized image result, FIG. 9B is a diagram illustrating a labeling result, and FIG. 10 is a graph showing values of parameters used for adaptive threshold determination for obtaining a threshold.

The labeling process separates adjacent pixels connected in eight directions into the same area and separates them into independent blob units. At this time, in the preferred embodiment of the present invention, if the number of pixels constituting the independent blob is less than 20, it is regarded as an unnecessary blob and the system may not automatically consider it. Such criteria may be appropriately selected by those skilled in the art to which the present invention pertains. As shown in FIG. 9B, it can be confirmed that many small noises shown in FIG. 9A are canceled out. The blobs left after removing the noise are called B ₀ and the subscript O is defined as the index of the blobs.

The lanes have a color and a long rectangular shape in one direction that distinguishes them from other objects in the image. In a preferred embodiment of the present invention, the feature may be defined using the ratio of the width to the length of the lane. This ratio is calculated from the eigenvalues obtained from the covariance matrix of B ₀ , and the eigenvectors corresponding to the large eigenvalues are designated as the principal major axis P ₀ of the blob. Equation 9 below is used to compare two eigenvalues to find the ratio M ₀ of the principal axis.

And as shown in Figure 9b the geometric modification of the lanes exist in a diagonal line in the direction of the upper left to the lower right of the image. In a preferred embodiment of the present invention, only blobs in which the angle A ₀ of the main main axis P ₀ of the lane is larger than 0 degrees and smaller than 50 degrees can be considered. The angle A ₀ is represented by equation (10). Represents a blob F ₁ B ₀ of interest for satisfying the above two conditions represented by the equation (8) may be used in the following process to obtain the vanishing point of interest only blobs F _1. The angle A ₀ is calculated by the equation (10). P ^x ₀ and P ^y ₀ are the lengths of the horizontal x-axis and the vertical y-axis in the image of the main principal axis.

The intersection analysis of the blobs creates an equation of the straight line of the major principal axis using the principal major axis of the blob F ₁ of interest. The intersection point X _u (x, y) is obtained by considering all the major axes of F ₁ . 11 is a graph representing the intersection point X _u (x, y) thus obtained.

를 기준으로 상기 반지름 이내의 영역 내 교차점의 무게 중심을 계산하고 이를 소실점 V(x,y)으로 결정한다.According to a preferred embodiment of the present invention, the center of the highest density small region may be determined as the vanishing point by analyzing the distribution of the intersection points X _u (x, y) obtained above. In order to determine the highest density region, the distance between the reference intersection point and the intersection point within the area is calculated by using the intersection point in the area existing on a circle having a constant radius centered on an arbitrary reference point. The index u of the reference intersection point at which the sum of the intersection points in the area for all the reference intersection points is maximum is determined as h.

The center of gravity of the intersection point in the area within the radius is calculated and determined as the vanishing point V (x, y).

12 is a graph illustrating intersection points within a circle (the highest density region in FIG. 10) when the radius is 20 according to an exemplary embodiment of the present invention, where A is a reference of the region determined as the highest density region. The intersection X _h (x, y), B is the center of gravity of the intersection in the region determined by the vanishing point V (x, y), and the black point represents the intersection in the region of X _u (x, y).

A vanishing line is created by connecting the vanishing point V (x, y) defined above to the center of gravity of the blob F ₁ of interest. The system selects a lane boundary candidate having an average angle of corresponding vanishing lines if the generated vanishing lines are within 5 degrees. This process is applied to all vanishing lines to obtain all lane line candidates. The solid black line in FIG. 13A shows all the disappearance lines generated in this process.

Multiple lane boundary candidates are obtained as a result of this process, but not all detected candidates need to be used. As described above, the proposed algorithm considers a situation in which a driving car changes lanes to a side lane. Therefore, among the obtained lane boundary candidates, two lane boundary candidates near the left and bottom corners of the image are set as lane boundary lines. The solid black line in Fig. 13B represents a lane boundary.

FIG. 13C is a diagram illustrating a currently running lane and a lane to be entered when the lane is changed, and a closed surface drawn based on the left and bottom corners and the first lane boundary line from the lane boundary line detected above (area indicated by ① in FIG. 13). In the present invention, it is recognized as the area of the currently driving lane. In addition, in the present invention, a closed surface (area indicated by ② in Fig. 13) drawn on the basis of the first and second lane boundary lines may be recognized as a lane to be entered when the lane is changed.

In the lane change estimation time calculating step (S030) according to an exemplary embodiment of the present invention, the lane change direction, the lane change start time, and the lane change completion time are calculated by calculating the position, angle, and moving speed of the lane boundary. The expected time can be calculated.

FIG. 4 is a control flowchart illustrating the lane change expected time calculating step of FIG. 1 in more detail. As shown, the estimated lane change time calculation step (S030) according to an embodiment of the present invention; Detecting a vehicle boundary line from an input image (S031); Determining a lane change direction by calculating an angle of a vehicle boundary line overlapping the lane boundary line (S032); Comparing an overlapping position and a degree of the vehicle boundary line and the lane boundary line with a reference value stored in advance and determining a starting point of a lane change when the reference value overlaps with the reference value or more (S033); Detecting a size and a moving speed of the vehicle from previously stored vehicle data, and predicting a lane change completion time when the overlapping position exceeds a reference value by comparing with the overlapping position and the degree of the boundary line (S034); And calculating and displaying the expected lane change time (S035).

14 is a block diagram illustrating an embodiment of a lane recognition apparatus using road area and geometric information according to the present invention.

As shown, an embodiment of the apparatus according to the present invention obtains a surface feature standard vector for a region of interest from a plurality of road images stored in the DB 40 and compares the road region with a surface feature vector of the input image. Road area determination means (10) for determining; Boundary line and lane detection means 20 for detecting an intersection with a lane using geometric information of the lane within a road area, acquiring a vanishing point, and recognizing a lane boundary and a lane; And a lane change predicted time calculating means 30 for determining the lane change direction, the lane change start time and the lane change complete time by calculating the position, the angle, and the moving speed of the lane boundary, and calculating the estimated time for the lane change direction.

The road area determining means 10 sets a region of interest according to the same distance in any one specific image in a DB including a plurality of road images; Divide the region of interest into a plurality of partitions, and obtain respective surface feature vectors for each partition; Calculating a mean for each surface feature vector to obtain a surface feature standard vector for the region of interest; A surface feature vector of the ROI is calculated from the input image and compared with the surface feature standard vector to determine a road region corresponding to the ROI. At this time, the object for obtaining the surface feature vector is preferably any one or more data selected from image energy, contrast, homogeneity and correlation.

The boundary line and lane detection means 20 obtain an adaptive threshold value from the road area; Binarize the image; Perform labeling using a Pixel coefficient and store the blob of interest; Detecting a lane using lane individual features (axis information); Acquire intersection points using key principal information of the blob of interest; Determine the highest density small region through intersection analysis; The vanishing point is obtained by the center of gravity of the small region; Establishing lane boundary candidates through vanishing line acquisition and angle integration; Following the vanishing point and the lane, the lane boundary line according to the position is detected and the lane is recognized through the boundary line.

The lane change estimated time calculating means 30 detects a vehicle boundary line from an input image, calculates an angle of the vehicle boundary line overlapping the lane boundary line, determines a lane change direction, and overlaps the position of the vehicle boundary line with the lane boundary line. When the degree is detected and overlapped by a reference value or more, the start point of the lane change is determined, the size and the moving speed of the vehicle are detected, the lane change completion time is predicted, and the estimated lane change time is calculated and displayed.

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 embodiment, and that various changes and 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.

10: road area determination means
20: boundary line and lane detection means
30: estimated lane change time
40: Database

Claims (6)

A road region determination step of obtaining a surface feature standard vector of a region of interest from a plurality of road images stored in a DB, and determining a road region by comparing the surface feature vector of an input image;
A boundary line and a lane detection step of detecting an intersection with a lane using geometric information of the lane within the road area, acquiring a vanishing point, and recognizing a lane boundary and a lane; And
A lane change prediction time calculating step of determining a lane change direction, a lane change start time and a lane change completion time by calculating a position, an angle, and a moving speed of a lane boundary, and calculating an estimated time for the lane change direction;
The road area determination step,
Setting a region of interest according to the same distance in any one specific image in a DB including a plurality of road images;
Dividing the region of interest into a plurality of partitions and obtaining respective surface feature vectors for each partition;
Calculating a mean for each surface feature vector to obtain a surface feature standard vector for the region of interest;
Calculating a surface feature vector of the divided region in the input image and comparing the surface feature standard vector included in the region of interest according to the corresponding distance with the region to determine the road region; Recognition method by difference using geometric information.
delete The method of claim 1,
The object for obtaining the surface feature vector of the road area determination step,
The road recognition method using the road area and geometric information, characterized in that any one or more data selected from the image energy, contrast, homogeneity and correlation.
The method of claim 1,
The boundary line and lane detection step;
Obtaining an adaptive threshold value from the road area;
Binarizing the image of the road area;
Performing labeling using a pixel coefficient and storing a blob of interest;
Detecting a lane using lane individual features (spindle information);
Acquiring the intersection using the main axis information of the blob of interest;
Determining the highest density small region through intersection analysis;
Obtaining a vanishing point based on the center of gravity of the small region;
Establishing a lane boundary candidate through vanishing line acquisition and angle integration; And
Detecting a lane boundary line according to a location after the vanishing point and the lane, and recognizing the lane through the boundary line.
5. The method of claim 4,
The lane change estimated time calculating step;
Detecting a vehicle boundary line from an input image;
Determining a lane change direction by calculating an angle of a vehicle boundary line overlapping the lane boundary line;
Comparing an overlapping position and a degree of the vehicle boundary line with the lane boundary line with a reference value stored in advance and determining a starting point of a lane change when the reference value overlaps with the reference value;
Detecting a size and a moving speed of the vehicle from previously stored vehicle data, and predicting a lane change completion time when the overlapping position exceeds a reference value by comparing the overlapping position and the degree of the boundary line; And
Calculating and displaying the estimated lane change time; and using the road area and the geometric information.
Road area determination means for obtaining a surface feature standard vector of the ROI from a plurality of road images stored in a DB, and determining a road area by comparing with a surface feature vector of the input image;
Boundary lines and lane detection means for detecting intersections with lanes using geometric information of lanes within the road area, acquiring vanishing points, and recognizing lane boundary lines and lanes; And
A lane change estimation time calculating means for calculating a position, an angle, and a moving speed of a lane boundary to determine a lane change direction, a lane change start time, and a lane change completion time, and calculate an estimated time for the lane change direction;
The road area determination means,
Setting a region of interest according to the same distance in any one specific image in a DB including a plurality of road images; Divide the region of interest into a plurality of partitions and obtain respective surface feature vectors for each partition; Calculating a mean for each surface feature vector to obtain a surface feature standard vector for the region of interest; A road using a road area and geometric information is calculated by calculating a surface feature vector of a segmented area from an input image, and determining a road area by comparing the surface feature standard vector included in the region of interest according to the corresponding area. Recognition device.

Images