KR101392850B1 - Method and system for lane departure warning based on image recognition - Google Patents

Abstract

The present invention relates to a lane departure detection method and system, and an object of the present invention is to provide a lane departure detection method based on image recognition, capable of maximizing lane recognition accuracy by being robust to various road conditions such as a lane departure, various noises, System.
The lane departure detection method based on the image recognition according to the present invention includes a first step of setting a region to be used for image processing for detecting a lane marker in an image photographed through a camera of a vehicle Wow; A second step of dividing and setting the entire ROI into a plurality of ROIs (hereinafter referred to as a ROI); A third step of extracting an edge per unit ROI including a lane marker using an edge detection technique; A fourth step of extracting a straight line through a Hough Transform using the edge per unit ROI extracted in the third step; A fifth step of extracting a plurality of point components corresponding to the straight line and spaced apart from each other on the straight line extracted in the fourth step; A sixth step of defining the lane model by applying the plurality of point components extracted in the fifth step to the least squares method; And a seventh step of determining whether or not the lane departure is determined by using the lane model defined in the sixth step.

Description

TECHNICAL FIELD [0001] The present invention relates to a lane departure detection method and system based on image recognition,

The present invention relates to a lane departure detection method and system, and more particularly, to a lane departure detection method and system that divides an image input from a camera mounted on a vehicle into a unit area of interest, extracts a point component, And more particularly, to a method and system for detecting lane departure based on image recognition.

In general, the lane departure warning system is configured to determine whether or not the lane departure of the vehicle is caused by using the image information photographed by the camera, and to perform a predetermined alarm process when it is determined that the lane has been deviated from the lane.

The conventional lane departure warning system is configured to recognize a lane in a line form and judge whether or not the lane is departed. The conventional lane departure recognition method is a method of recognizing a lane departure warning system in which a lane width, a radius of curvature of a lane, It is difficult to recognize the lane departure exactly because it includes as many noise sources as many various variables such as the difference between the optical axes must be considered.

In addition, the conventional lane departure warning system is inaccurate when a sign other than a lane marker is formed on a road or a road where the lane marker is cut off in the middle. Accordingly, even when the vehicle does not depart from the center of the lane in operation, There is a problem that there are elements that can generate sound.

Korean Patent Laid-Open Publication No. 2000-0037604 discloses a method for recognizing a lane. According to this technique, two cameras are used to recognize a lane, and a portion having a higher luminance than a road surface portion is recognized as a lane. However, in general, in addition to lanes, there may be other foreign objects such as fallen objects falling on a vehicle loading floor, etc. In the case where such foreign objects have a higher brightness than the road surface, errors may be recognized as lanes. In addition, the use of two cameras increases the manufacturing cost and requires a large installation space, which is a very difficult problem to implement as a compact device.

Also, in the conventional lane recognition method, only the Sobel filter is used to extract the lane edge and the lane marker is recognized through this method. However, this method is very vulnerable to noises, so that there are many errors in recognizing the lane that is not an actual lane.

SUMMARY OF THE INVENTION The present invention has been conceived to solve the above-mentioned problems, and it is an object of the present invention to provide an image processing apparatus and a method for preventing a lane-finding error caused when a lane is broken, a sign other than a lane is present, A lane departure detection method and system based on image recognition capable of accurately recognizing a lane in a situation such as a curved road while effectively eliminating noise with only one camera.

According to another aspect of the present invention, there is provided a method of detecting a lane departure based on image recognition, the method including detecting a lane mark in an image captured through a camera of a vehicle, The method comprising: A second step of dividing and setting the entire ROI into a plurality of ROIs (hereinafter referred to as a ROI); A third step of extracting an edge per unit ROI including a lane marker using an edge detection technique; A fourth step of extracting a straight line through a Hough Transform using the edge per unit ROI extracted in the third step; A fifth step of extracting a plurality of point components corresponding to the straight line and spaced apart from each other on the straight line extracted in the fourth step; A sixth step of defining the lane model by applying the plurality of point components extracted in the fifth step to the least squares method; And a seventh step of determining whether or not the lane departure is determined by using the lane model defined in the sixth step.

According to the image recognition-based lane departure detection method and system according to the present invention, a camera input image is divided and processed into a unit area of interest, so that road information (milestones, etc.) existing between left and right lane markers, It is possible to generate a virtual lane that matches the actual lane as much as possible even in the case of a lane having an area cut in the middle, thereby remarkably improving the lane recognition accuracy.

In addition, since the entire region of interest is divided into the vertical and horizontal directions to process only a part of the region (i.e., the ROI), it is possible to minimize an actual image processing region, thereby reducing the computational load of the image processing.

In addition, it is possible to increase the accuracy by excluding edges not included in a straight line by applying representative point components from the straight line extracted through Hough transformation using edge information instead of applying the edge points of the lane to the least squares method, There is an advantage of reducing the computation load by reducing the number of points to use in the square law.

Brief Description of the Drawings Fig. 1 is an overall configuration diagram of a lane departure detection system based on image recognition according to the present invention; Fig.
FIG. 2 is a block diagram showing a detailed configuration of the image recognition processing unit shown in FIG. 1. FIG.
FIG. 3 is a coordinate conceptual diagram of a lane departure detection method and system based on image recognition of the present invention.
4 is a block flow diagram illustrating a process flow of a lane departure detection method based on image recognition according to the present invention.
5 is an example of a whole region of interest set through the location of the disappearance line and the hood of the present invention.
6 (a) is an example of a general black-and-white image.
6 (b) is an example of a monochrome image in which a yellow component is emphasized by the image conversion unit of the present invention.
7 (a) is an example of a unit interest area set when a lane marker is not recognized in a previous frame.
7 (b) is an example of a unit interest area set when a lane mark is recognized in a previous frame.
8 is an example of a horizontal slope image obtained by the Sobel operation of the edge extracting unit of the present invention.
9 is an example of a tilt magnitude image obtained by the Sobel operation of the edge extracting unit of the present invention.
10 is an example of an edge image obtained through the edge extracting unit of the present invention.
11 is an example of a straight line obtained through the linear-information extracting unit of the present invention.
12 is an example of dot components extracted by the dot extracting section of the present invention on a straight lane;
13 is an example of a dot component extracted from a curve lane of the dot extracting part of the present invention.

The present invention is capable of eliminating various noise sources as much as possible and further accurately recognizing the lane marker without recognizing errors even under various environmental conditions such as various road conditions such as a straight road and curved road, Disclose technical features that can alert.

In the following, preferred embodiments, advantages and features of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is an overall configuration diagram of a lane departure detection system based on image recognition according to the present invention, and FIG. 2 is a block diagram showing a detailed configuration of an image recognition processor shown in FIG.

1 and 2, the image recognition based lane departure detection system of the present invention is configured to include a camera 10, an image recognition processing unit 20, and an alarm output unit 95.

The camera 10 of the present invention is a component mounted on a vehicle and capturing an image by photographing a running road ahead of the vehicle, preferably a CMOS camera or a CCD camera.

The image recognition processing unit 20 recognizes a lane indicated on the image acquired by the camera 10 and performs a predetermined calculation process based on the lane indicated on the image, A function to prevent accidents caused by driving or carelessness is realized.

The image dividing unit 40, the edge extracting unit 50, the linear information extracting unit 60, the dot extracting unit 70, A lane model generation unit 80, and a lane departure determination unit 90. [

The lane departure detection system based on the image recognition of the present invention first initializes the system in order to realize the lane departure detection function by recognizing and processing the image obtained through the camera in the running process of the vehicle.

In the system initialization, an area (hereinafter, referred to as an 'entire ROI') to be used for image processing for detecting a lane marker on an image photographed through a camera of a vehicle is set, a memory for processing various images is allocated, Information about the mark and road width, information on the position of the front wheel of the vehicle, and information necessary for performing the Hough Transform are stored in a lookup table.

Upon completion of the system initialization, recognition processing of the input image and lane departure detection based thereon are performed through the following operations of the respective components of the image recognition processor 20.

The image converting unit 30 of the present invention basically converts the image input from the vehicle camera into a monochrome image, and preferably performs a process of emphasizing the yellow color component in the converted monochrome image.

The image segmentation unit 40 of the present invention divides the entire ROI set in the monochrome image into a plurality of regions (hereinafter, referred to as unit ROIs). Wherein the unit interest area is a rectangular area formed by dividing the entire ROI into a longitudinal direction and a lateral direction, and the rectangular ROI includes a left lane marker and a right lane marker, It is preferable to configure them so as to be continuously arranged and arranged in a row.

The edge extracting unit 50 of the present invention extracts edge information for each ROI including a lane mark using an edge detection technique. Here, the edge detection technique may employ a Sobel edge extraction method for detecting a boundary (contour) of objects in an image using a change in brightness value by a differential operator.

The linear information extracting unit 60 of the present invention extracts straight line information through a Hough Transform using the edge per unit area of interest extracted by the edge extracting unit 50. [

For reference, the Hough Transform used in the operation of the linear-information extracting unit 60 is the most widely used method for finding a straight line in the image processing. It converts a linear equation in a two-dimensional image coordinate system into a parameter space To find a straight line. Since Hough transform is a commonly used algorithm in engineering, a detailed description is omitted.

In addition, when there is a unit interest region (hereinafter, referred to as a 'non-extraction unit interest region') in which a straight line can not be extracted, the linear information extraction unit 60 extracts a unit Extracts a straight line to be included in the non-extraction unit ROI through analogy using the linear information included in the ROI, and thereby makes it possible to derive the complete lane marking successively.

The dot extracting unit 70 of the present invention extracts a plurality of point components corresponding to the straight line and spaced apart from each other on a straight line extracted by the straight line information extracting unit 60. [ Accordingly, one of the plurality of dot components extracted by the dot extracting unit 70 is aligned with the left lane marking and arranged at an interval with the left lane, and the other extraction group includes the right lane marker And they are arranged in the right lane with an interval between them.

The lane model generating unit 80 of the present invention functions to define a lane model by applying a plurality of point components extracted by the dot extracting unit 70 to the least squares method.

More specifically, the lane model generating unit 80 determines whether the lane markers displayed on the corresponding image are straight lines or curves using the slopes of the straight lines extracted for each unit area of interest, and if the determination result is a straight line, The lane marking model is defined by applying an approximate least squares method and the lane marking model is defined by applying a least squares method approximating a quadratic expression to the determination result curve to determine a reference line (i.e., lane marking) to be used for lane departure detection.

The lane departure determination unit of the present invention determines whether or not the lane departure of the vehicle is caused by using the lane model determined by the lane model generation unit 80 (i.e., the reference line used for lane departure detection). Specifically, And if the distance calculated by the calculation exceeds the reference value, it is determined that the lane departure is not possible and the alarm output unit 95 is provided with an alarm generation signal.

On the other hand, the lane departure determination unit may be configured to provide the alarm output unit 95 with another advance alarm generation signal corresponding to not only when the vehicle departs from the lane of the vehicle, but also when the vehicle approaches the lane of a certain distance or less to be.

The alarm output unit 95 of the present invention outputs a visual or audible alarm to the outside according to an alarm generation signal or a pre-alarm generation signal sent from the lane departure determination unit so that the driver can recognize the fact.

It is preferable that the alarm according to the alarm generation signal and the alarm according to the advance alarm generation signal are configured to be outputted in different forms so as to guide the driver to distinguish whether the vehicle has just left the vehicle or has already departed.

Hereinafter, functions and actions performed by each component of the image recognition processor 20 for image recognition and processing will be described in detail.

3 is a coordinate conceptual diagram of a lane departure detection method and system based on image recognition of the present invention. For reference, the camera 10 of FIG. 3 is mounted on a vehicle (not shown) and can use a CMOS or CCD camera.

3, X _W , Y _W , Z _W Means a world coordinate system which covers the whole running road space, X _C , Y _C , Z _C Means a coordinate system of a camera image frame centered on the camera viewpoint. The camera is installed so as to be inclined at an angle? With respect to the road surface on which the lane marker is formed, and the height from the road surface is indicated by H.

The lane departure detection method and system according to the present invention is used in a lane departure detection method and a lane departure determination method, in which information such as a lane mark thickness and a lane width on a world coordinate system are converted into values on an image, The relationship of the image coordinate system will be described.

The world coordinate system (X _W , Y _W , Z _W ) and the camera coordinate system (X _C , Y _C , Z _C ) are transformed into a plane image (1) < / RTI >

(In the equation (1)

X _C , Y _C , Z _C : X, Y, Z coordinates of the camera coordinate system,

X _W , Y _W , Z _W : X, Y, Z coordinates of the world coordinate system,

[theta]: an angle at which the camera optical axis Z _C is inclined with respect to the horizon line Y _W ,

H: height from the road surface to the position where the camera is installed)

Using the pinhole camera model and Equation (1), the image coordinate system can be expressed by the following Equation (2).

(In the equation (2)

u: the horizontal axis coordinate of the image coordinate system,

v: the vertical axis coordinates of the image coordinate system,

f: the focal length of the camera)

On the other hand, since the lane marking to be recognized exists on the road surface, the value of Z _W can be assumed to be '0'. Thus, if Z _W = 0 is substituted into Equation 2, the image coordinate system is simplified.

Using the horizontal and vertical resolutions of the image sensor and the horizontal and vertical sizes of the pixels, the coordinates on the pixel coordinate system can be expressed by the following Equation (3).

(In Equation 3,

x _P , y _P : Horizontal, vertical axis pixel coordinates,

r _w , r _h : Horizontal and vertical resolution of the image sensor,

s _u , s _v : the horizontal and vertical pixel sizes of the image sensor)

Hereinafter, a method for recognizing a lane departure by recognizing a lane marker on an input image based on the relationship between the world coordinate system, camera, and image coordinate system as described above will be described in detail.

4 is a block flow diagram illustrating a flow of a lane departure detection method based on image recognition according to the present invention.

Referring to FIG. 4, an image recognition based lane departure detection method according to the present invention is a method for detecting a lane mark in an image captured through a camera of a vehicle, (Step 1); Dividing the entire ROI into a plurality of regions (hereinafter, referred to as 'ROI') (Step 2); (Step 3) of extracting an edge for each ROI including a lane marker using an edge detection technique; Extracting a straight line through a Hough Transform using the edge per unit RO extracted in Step 3 (Step 4); Extracting a plurality of point components corresponding to the straight line and spaced apart from each other on the straight line extracted in step 4; Applying the plurality of point components extracted in step 5 to the least squares method to define a lane model (step 6); And determining whether the lane departure is determined using the lane model defined in step 6 (step 7).

Preferably, step 1 may further comprise the step of converting the image input from the vehicle camera into a monochrome image, and a step of enhancing the yellow color component in the monochrome image (step 1-1).

The unit interest area in step 2 is formed of a rectangular area formed by dividing the entire ROI into a longitudinal direction and a lateral direction, and the rectangular ROI is divided into a left lane marker and a right lane marker And is arranged so as to be continuous and continuous along the lane marker.

The edge detection technique of step 3 may preferably use a Sobel Edge extraction scheme.

Step 4 may further include a step of obtaining a slope average value of a straight line extracted for each unit ROI and a step of removing a straight line having a difference of more than a predetermined angle from the slope average value of the extracted ROIs.

In the lane departure detection method of the present invention, when there is a unit interest area (hereinafter, referred to as 'non-extraction unit interest area') in which a straight line is not extracted in step 4, And generating an imaginary straight line to be included in the non-extracted unit ROI through analogy using linear information included in the ROI.

Here, the inference using the linear information included in the unit interest area arranged before and after means that a straight line connecting the lower end and the upper end of the straight line included in the unit interest area arranged before and after is generated, A method of generating a straight line corresponding to this can be used.

The method may further include determining whether a lane marker of the image photographed through the camera is a straight line or a curve using the slope of the straight line extracted for each interest area through step 4 In this case, step 6 defines a lane model by applying a least squares method approximating a linear equation if the determination result is a straight line, and defines a lane model by applying a least squares method approximating a quadratic equation if the determined result curve is a curve do.

Hereinafter, an example according to each step will be described in more detail.

≪ Step 1 (S10)

Step 1 of the present invention is a step (S10) of performing initialization in order to realize a lane departure detection function by recognizing and processing an image obtained through a camera in a running process of a vehicle, using Equations 1 to 3 Find the ordinate of the missing line on the image. The disappearance line means a line corresponding to a point (vanishing point) at which it meets with the horizon among a plurality of lines dividing in the horizontal direction along the vertical axis of the entire area of interest.

The ordinate of the disappearance line can be calculated by Equation (4) as Y _W = Y _P when the Y _W of the world coordinate system converges to infinity (∞).

In the image, an entire region of interest is set by setting the ordinate value of the disappearance line to the upper end and the ordinate value of the position where the vehicle hood is located to the lower end. The line between 'vanish' and 'hood' in FIG. 5 corresponds to the entire region of interest to be used for image processing.

Step 1 of the present invention performs the following initialization in addition to setting the entire ROI. First, a memory for image processing such as a monochrome image, a horizontal gradient image, and a Hough Transform is allocated.

Then, sin and cos values necessary for Huff transform are calculated and stored in a lookup table.

The minimum and maximum thicknesses of the lane marker on the world coordinate system and the minimum and maximum widths of the road on the world coordinate system are converted into values corresponding to all the ordinates corresponding to the upper and lower ends of the entire ROI on the image using the above- Lookup table.

Then, in order to determine whether or not the vehicle departs from the lane using the coordinate system formulas, an ordinate corresponding to the front wheel position on the image is calculated and stored.

≪ Step 1-1 (S15)

Step 1-1 of the present invention is performed by the image converting unit 30. Specifically, after the image input from the vehicle camera is converted into a monochrome image, the Cb of the YCbCr color space Emphasize the yellow component in the monochrome image by using the color component as follows.

{IF Cb -128 < 0 THEN

    Gray Pixel = Y + (128 - Cb) * 2

ELSE

    Gray Pixel = Y}

FIG. 6 (a) shows a general black-and-white image, and FIG. 6 (b) shows a black-and-white image in which the yellow component is emphasized in step 1-1. As can be seen from FIG. 6, it can be seen that the lane marker is more clearly contrasted in the monochrome image by emphasizing the yellow component using the Cb color component.

&Lt; Step 2 (S20)

Step 2 of the present invention is performed by the image segmenting unit 40 as a step of setting a unit ROI corresponding to the left and right lane markers among the ROIs of the image.

The lane marking is characterized by a sharp, linear component stronger near the vehicle, blurred away from the vehicle, and a straight line component weaker in the case of a curved road. Accordingly, as shown in FIG. 5, a plurality of horizontal lines 41 spaced apart along the vertical axis divide the entire area of interest into an appropriate number so that the lane marker recognition can be processed for each divided area.

Thus, if the whole area of interest is divided and processed along the vertical axis, it is more robust in lane marker recognition and straight line extraction in a situation like a curved road. Also, even if the lane marker recognition is difficult at a specific point (unit interest area) as in the case of the dotted line lane marker or the lane marker being damaged, the lane marker information of the specific point can be inferred There is an advantage to be able to do.

As described above, the ROI divided into a plurality of lines along the vertical axis is divided into a plurality of vertical lines 42 along the horizontal axis by the left and right lanes, and finally, a unit ROI including a rectangular area is set as shown in FIG. 7 .

Here, it is preferable that the rectangle unit ROI is configured to be continuously arranged and arranged along the left lane markers and the right lane markers along the corresponding lane markers.

On the other hand, the width of each unit area of interest is set differently depending on whether the lane marker is recognized in the previous frame or not.

FIG. 7 (a) shows a set unit interest area when the lane marker is not recognized in the previous frame, and FIG. 7 (b) shows the set unit interest area when the lane marker is recognized in the previous frame.

First, when the lane marker is not recognized in the previous frame, the width between the minimum width and the maximum width of the road is set as the width of the unit interest area on the basis of the center of the image on the horizontal line of the unit interest area. ) Reference)

Next, when the lane marker is recognized in the previous frame, the lane marker is recognized as a center of the lane marker recognized in the previous frame, and is expanded to the left and right by a predetermined value to set the horizontal width of the unit interest area (refer to FIG. 7B) )

By limiting the width of the unit area of interest as described above, there is an advantage that various noise elements that can be mistaken for lane markings or road information (milestones etc.) existing between the left and right lane markers can be removed.

In addition, since the entire region of interest is divided into the vertical and horizontal directions to process only a part of the region (i.e., the ROI), it is possible to minimize an actual image processing region, thereby reducing the computational load of the image processing.

&Lt; Step 3 (S30) >

Step 3 of the present invention is performed by the edge extracting unit 50 as a step of extracting an edge per unit ROI including a lane mark using an edge detection technique.

To be more specific, Sobel operation is performed for lane marker recognition to obtain a horizontal slope image and a slope image.

The mask for the Sobel operation is as follows. Gx denotes a horizontal component, and Gy denotes a vertical component.

The horizontal slope image is a horizontal component of the mask for the Sobel operation, and FIG. 8 shows an example of such a horizontal slope image.

The tilt magnitude image is obtained by using the following equation (5) using the horizontal component and the vertical component.

FIG. 9 shows an example of a tilt magnitude image calculated through the above method.

In order to recognize the lane marker using the black and white image, the horizontal gradient image, and the tilt magnitude image obtained in steps 1-1 and 3, the following lane marker characteristics are used.

- The brightness of the center area is brighter than the brightness of the left and right area based on the left and right boundaries of the lane marker.

- The brightness change of the lane marker is brightened from left to right and has a characteristic of darkening. Is the same as the portion indicated by each point 51a and 51b at the point 51a where the white pixel of FIG. 8 becomes bright and the point 51b where the black pixel becomes dark.

- The left and right boundaries of the lane marker are the points having the local maximum value in the tilt magnitude image. The portion indicated by the dot (red dot) 52 in FIG. 9 is the left and right boundary portion of the lane marker having the local maximum value.

In the horizontal slope image, the inner boundary of the left lane marker is a point having a negative value, and the inner boundary of the right lane marker is a point having a positive value.

Using the above four lane marking characteristics, the lane edge image is obtained as follows.

a) Set the ordinate of the top at y and the ordinate at the bottom at y _max in each region of interest.

b) A pixel x ₁ having the local maximum value of the gradient magnitude image and moving in the right direction starting from the left side of the image corresponding to the y coordinate of the ROI and having a positive value in the horizontal gradient image is searched.

c) Again, move to the right with respect to x ₁ and find the pixel x ₂ with the local maximum of the tilt magnitude image and the negative of the horizontal gradient image.

d) the interval between x ₁ and x _{2 is} between the minimum and maximum values corresponding to the y coordinate of the lane mark thickness lookup table on the image obtained in the initialization step, and the distance between the x ₁ and x ₂ points of the monochrome image If the average of the brightness values is greater than the brightness values of the pixels (x ₁ -1, x ₂ +1) adjacent to x ₁ and x ₂ , it is determined as a lane marker and stored as an edge of the difference marker. At this time, x ₂ is stored as an edge pixel in case of a unit interest area corresponding to the left lane, and x ₁ is stored as an edge pixel in case of a unit interest area corresponding to the right lane.

e) Repeat steps c to d for all pixels in the ROI corresponding to the y coordinate, and repeat steps b to d again by increasing y by 1 when all processing is completed. This process repeats until y reaches y _max .

10 is an example of an edge image of a lane marker obtained by the above method. The point (red dot) in Fig. 10 corresponds to the right point among the two points shown in Figs. 8 and 9, respectively.

&Lt; Step 4 (S40)

Step 4 of the present invention is a step of obtaining a parameter (?,?) For a straight line by performing a Hough Transform using edge information for each ROI extracted in Step 3, , Where the straight line is expressed by the following Equation (6).

On the other hand, it is necessary to work to remove a straight line generated by a noise component in a straight line extracted by the Hough transform and not coinciding with a lane marking. The straight line by the noise component can be removed by the following method.

That is, the straight line of the unit interest area at the current stage removes the disconnected straight line without continuing to each other by using the characteristic of connecting the straight line of the unit interest area in the pre and post stages. Also, a straight line having a slope difference of more than a certain angle from the slope mean value among straight lines of the unit area of interest extracted through the Hough transform in step 4 is removed.

On the other hand, when there is a unit interest area (non-extraction unit interest area) in which a dotted line lane marker or lane marker is damaged or a straight line can not be extracted due to a noise component, Extracted unit interest region through the analogy using the straight line information included in the unit interest region.

11 shows a straight line obtained by performing Hough Transform using edge information. Meanwhile, in FIG. 11, the unit interest region set in the third order of the right lane corresponds to the non-extraction unit interest region, and the non-extraction unit interest region is an inference using the linear information included in the pre- A virtual straight line can be generated.

&Lt; Step 5 (S50)

Step 5 of the present invention is performed by the dot extracting unit 70 as a step of extracting a plurality of dot components corresponding to the straight line and arranged mutually spaced from each other on the straight line obtained in step 4. [

One of the plurality of dot components extracted by the dot extracting unit 70 is aligned with the left lane marker and arranged at a distance from each other along the left lane, and the other extract group is aligned with the right lane marker And they are arranged in the right lane with an interval between them.

Fig. 12 shows the point component 71 extracted by step 5 for the straight lane, and Fig. 13 shows the point component 72 extracted by step 5 for the curve lane.

The plurality of point components thus extracted can be used in the least squares method of step 6 to define the lane model specification.

&Lt; Step 6 (S60)

Step 6 of the present invention is a step of defining the lane model by applying the plurality of point components extracted in step 5 to the least squares method, and is performed through the lane model generation unit 80. [

The lane model generating unit 80 defines a lane model by applying a least squares method approximating a linear equation (x = a + by) if the corresponding point component is a point component for a linear lane marking, (X = a + by + cy ² ), the lane model is defined by applying the least square method approximating the quadratic expression (x = a + by + cy ² ) to determine a reference line (i.e., lane marker) to be used for lane departure detection.

Therefore, in order to define the lane model, before the point component is applied to the least squares method, the determination of whether the point component is a point component for a straight lane marking or a point component for a curved lane marking should be performed first.

A method of determining whether the lane marker is close to a straight line or near a curve is as follows. First, we calculate the slope average and average variance of the straight line per unit area of interest. If the calculated mean variance is smaller than the specific value and the slope mean value continuously decreases to the unit interest area set at the upper end of the left (or right) lane marker in the unit interest area set at the lower end of the left (or right) Or if it increases, it is judged as a curve, otherwise it is judged as a straight line.

In this case, if the lane edge points are not applied to the least squares method and instead, the representative point components extracted from the straight line extracted through the Hough transform using the edge information are applied, the edges not included in the straight line can be excluded and used for the least squares method There is an advantage of reducing the computation load by reducing the number of points.

&Lt; Step 7 (S70)

Step 7 is a step of judging whether or not the lane departure is determined by using the lane model defined in step 6 and is performed by the lane departure judging section 90. The distance to the front wheel of the vehicle and the lane marker is calculated to determine whether or not the lane departure do.

More specifically, in the initialization process of step 1, the ordinate in which the front wheel of the vehicle on the previously calculated image is located is substituted into the lane model equation to calculate the abscissa.

Then, the distance to the front wheel of the vehicle and the lane marker is calculated using the horizontal and vertical coordinates obtained by the calculation and the width of the vehicle. The numbers in green at the bottom of Figs. 12 and 13 are the distances to the front wheels of the vehicle and the lane marks.

That is, the distance between the left (or right) front wheel of the vehicle and the left (or right) lane marker is calculated, and if any of the distances calculated by the calculation is outside the reference value calculated based on the vehicle width, And provides an alarm generation signal to the alarm output unit 95. The alarm output unit 95 outputs an alarm sound or the like to the outside to warn the driver of the danger.

While the preferred embodiments of the present invention have been described and illustrated above using specific terms, such terms are used only for the purpose of clarifying the invention, and it is to be understood that the embodiment It will be obvious that various changes and modifications can be made without departing from the spirit and scope of the invention. Such modified embodiments should not be understood individually from the spirit and scope of the present invention, but should be regarded as being within the scope of the claims of the present invention.

10: camera 20: image recognition processor
30: image converting unit 40: image dividing unit
41: horizontal direction line of unit interest area 42: vertical direction line of unit interest area
50: edge extracting unit 60: straight line information extracting unit
70: dot extracting unit 71, 72: dot component
80: lane model generation unit 90: lane departure determination unit
95: Alarm output section

Claims (9)

A first step of setting an area to be used for image processing for detecting a lane marker in an image photographed through a camera of a vehicle (hereinafter, referred to as an 'entire ROI');
A second step of dividing and setting the entire ROI into a plurality of ROIs (hereinafter referred to as a ROI);
A third step of extracting an edge per unit ROI including a lane marker using an edge detection technique;
A fourth step of extracting a straight line through a Hough Transform using the edge per unit area of interest extracted in the third step;
A fifth step of extracting a plurality of point components corresponding to the straight line and spaced apart from each other on the straight line extracted in the fourth step;
A sixth step of defining the lane model by applying the plurality of point components extracted in the fifth step to the least squares method; And
And a seventh step of determining whether or not the lane departure has been made using the lane model defined in the sixth step.
The method according to claim 1,
In the first step,
And a 1-1 step of converting the image input from the vehicle camera into a monochrome image and then highlighting the yellow component in the converted monochrome image.
The method according to claim 1,
Wherein the unit interest area of the second step comprises a rectangular area formed by dividing the entire area of interest into a longitudinal direction and a lateral direction and the rectangular unit area of interest includes a left lane marker and a right lane marker, Wherein the lane departure detection method is configured to be continuously arranged and arranged along the lane marker for each landmark.
3. The method of claim 2,
Wherein the edge detection method of the third step is a Sobel edge extraction method.
The method of claim 3,
A fourth step of obtaining a slope average value of a straight line extracted for each unit ROI in the fourth step;
A fourth step of removing a straight line having a difference of more than a predetermined angle from the slope average value of the 4-1 step among straight lines of the unit ROI extracted in the fourth step; And
If there is a unit interest region (hereinafter, referred to as a 'non-extraction unit interest region') in which a straight line can not be extracted in the fourth step, the unit interest region included in the unit interest region arranged before and after the non- And generating a virtual straight line to be included in the non-extracted unit ROI through analogy using the straight line information.
The method of claim 3,
Further comprising the step of determining whether a lane marker of the image photographed by the camera is a straight line or a curve using the slope of the straight line extracted for each unit ROI through the fourth step,
The sixth step includes defining a lane model by applying a least squares method approximating a linear equation if the determination result is a straight line, and defining a lane model by applying a least squares method approximating a quadratic equation if the determination result curve is a curve Wherein the lane departure detection method is based on image recognition.
The method according to claim 6,
Wherein the step of determining whether the straight line or the curve is a curve,
Calculating a slope average and an average variance of a straight line for each unit ROI; And
If the calculated mean variance is greater than or equal to a specific value and the slope mean value continuously decreases or increases to a unit interest area set at the upper end of the left lane marker in the unit interest area set at the lower end of the left lane marking, The method comprising the steps of: (a)
The method according to claim 1,
In the seventh step,
Wherein the lane departure determination means calculates a distance between a left front wheel and a left lane marking of the vehicle and determines that the lane departure occurs when the distance calculated by the calculation exceeds a reference value calculated based on the vehicle width, Detection method of deviation.
A camera for photographing the front of the vehicle to acquire an image;
An image converter for converting the image input from the vehicle camera into a monochrome image;
Dividing the converted monochrome image into a plurality of regions (hereinafter, referred to as 'unit region of interest') and setting the divided regions;
An edge extracting unit for extracting an edge for each ROI including a lane marker using an edge detection technique;
A straight line information extracting unit for extracting a straight line through a Hough Transform using an edge for each unit area of interest extracted by the edge extracting unit;
A dot extracting unit for extracting a plurality of point components corresponding to the straight line and arranged at a distance from each other on a straight line extracted by the straight line information extracting unit;
A lane model generating unit for applying the plurality of point components to a least square method to define a lane model;
A lane departure determination unit for determining whether the vehicle is departing from the lane using the lane model; And
And an alarm output unit for outputting an alarm when the lane departure occurs as a result of the lane departure determination unit.

Images