KR20190103508A - Method for extracting driving lane, device and computer readable medium for performing the method - Google Patents

Abstract

According to the present invention, provided is a method for detecting a lane, which sets a first ROI estimated as a road area from a reverse projection converted image when a brightness value of an inputted road image is equal to or less than a threshold value, sets a second ROI estimated as a road area from the road image using a deep neural network, and detects a driving lane from a final ROI determined based on the first ROI and the second ROI, thereby searching a driving lane at constant performance regardless of the quality of the inputted road image.

Description

Lane detection method, apparatus and recording medium for performing the same {METHOD FOR EXTRACTING DRIVING LANE, DEVICE AND COMPUTER READABLE MEDIUM FOR PERFORMING THE METHOD}

The present invention relates to a lane detection method, an apparatus and a recording medium for performing the same, and more particularly, to a lane detection method for detecting a driving lane from a road image, an apparatus and a recording medium for performing the same.

BACKGROUND ART Due to the development of automobile technology, automobiles are being applied with technologies to ensure the safety of drivers and pedestrians and to improve the convenience of drivers. Examples of such technologies include adaptive cruise control and lane departure warning systems, and by combining and controlling these safety enhancement technologies, the vehicle is aware of the surroundings of the vehicle so that the vehicle itself can Control relevant functions and implement autonomous driving.

Furthermore, recently, technologies for recognizing and detecting an object on a road using a vision sensor have been developed, and pedestrians, vehicles, traffic lights, signs, and lane detection technologies are representatively developed. These technologies are being developed using image processing and machine learning of images acquired from cameras.

Object recognition technology using vision sensor is the most basic technology among ADAS that provides driving stability and convenience to the driver. In particular, vision sensors are inexpensive compared to other sensors and have excellent performance in recognizing the surrounding environment of vehicles. Therefore, vision sensors are essential for future autonomous vehicles. Vision sensor technology is being used in lane departure warning systems and autonomous emergency braking (AEB) as well as video storage devices such as black box systems. With the advancement of ADAS and autonomous driving technology, image processing technology using vision sensor is moving beyond development to the optimization and mass production stage.

However, the object detection technology using the conventional vision sensor is still limited to replace the human eye, and in particular, lane detection, an essential technology for determining the current position of the vehicle, preventing collision with other vehicles, and maintaining the position of the vehicle Is not functioning in the environment where the vision sensor cannot detect the lane, part or total loss due to lane aging, night driving and lighting problems. These problems may be fatal defects in autonomous vehicles in the future, so it is required to develop a system for detecting a lane even in a variety of environments or to estimate a lane and predict a path even if lane information is insufficient.

Korean Patent Registration No. 10-1240499 Korean Patent Publication No. 10-2014-0068475

One aspect of the present invention provides a lane detection method for detecting a driving lane using different methods according to the quality of a road image and driving conditions, and an apparatus and a recording medium for performing the same.

The technical problem of the present invention is not limited to the technical problem mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the following description.

In the lane detection method according to an embodiment of the present invention, extracting a brightness value of a road image, comparing the brightness value with a preset threshold value, estimated as a road area from a reverse projection converted image for the road image Setting a first ROI, and when it is determined that the brightness value is smaller than the threshold value, setting a second ROI including an object determined to be a lane and a road from the road image by using a pre-trained deep neural network; Determining a final region of interest estimated as a road region in the road image based on the first region of interest and the second region of interest and at least one driving from the final region of interest using a predetermined lane detection algorithm. Detecting a lane.

The setting of the second region of interest may include: labeling each pixel constituting the road image with any one of a pre-learned class by using the deep neural network, and classifying the pixels of the road image with respect to the learned class. It may be grouping into an object and setting pixels grouped into a road area class as the ROI.

The setting of the first ROI includes setting the leftmost solid line of the reverse projection converted image as a reference lane, determining a dotted line located within a predetermined reference distance from the reference lane as the lane, and setting the reference distance from the lane. One solid line located within the edge lane may be set as the edge lane, and the area from the reference lane to the edge lane may be detected as the first ROI.

The setting of the first ROI may further include determining a solid line located within a predetermined reference distance from the reference lane as a lane when the brightness value of the reverse projection converted image is smaller than a reference value, and continuously based on the reference distance. Among the determined lanes, the rightmost lane may be set as the edge lane, and the reference value may be set to a value smaller than the threshold.

The determining of the final region of interest may include mapping the second region of interest to the first region of interest so that the common region of the first region of interest and the second region of interest is based on the reference lane and the edge lane. Extracting a polygonal region may determine the final region of interest.

Extracting into the polygonal region may include setting the width of the final region of interest from the reference lane to the edge lane, and using the height of each lane divided by the lanes to determine the height of each polygonal section. It can be characterized by the setting.

Detecting the at least one driving lane converts any one point constituting a straight line of the final ROI into a straight line equation, and detects an intersection where a plurality of straight line equations converted for each point cross each other. The driving lane candidate group may be extracted using linear equations having a number of linear equations passing through the intersection point greater than or equal to a threshold.

The detecting of the at least one driving lane may include detecting the at least one driving lane from the driving lane candidate group using a RANSAC algorithm.

The determining of the final region of interest may include determining the detected first region of interest as the final region of interest when it is determined that the brightness value is smaller than the threshold value.

Before extracting the brightness value of the road image, the method may further include performing a preprocessing process of the road image, and performing the preprocessing process may include converting the road image into a gray scale image and using a top-hat filter. It may include converting the gray scale image using.

Further, it may be a computer-readable recording medium having a computer program recorded thereon for performing the lane detection method according to the present invention.

In addition, the lane detection apparatus according to an embodiment of the present invention, a driving environment determination unit for extracting the brightness value of the road image, and comparing the brightness value with a predetermined threshold value, from the reverse projection conversion image for the road image A first ROI setting unit configured to set a first ROI estimated as a road area, and when the brightness value is determined to be smaller than the threshold value, it is determined as a lane and a road from the road image by using a pre-trained deep neural network. A second region of interest setter configured to set a second region of interest including an object, and determine a final region of interest estimated as a road region in the road image based on the first region of interest and the second region of interest; At least one traveling lane is detected from the final region of interest using an area estimator and a predetermined lane detection algorithm. It includes a lane detecting unit.

The second ROI setting unit labels each pixel constituting the road image by using any one of a pre-learned class by using the deep neural network, and converts the pixels of the road image into an object for the learned class. The pixels grouped in the road region class may be set as the second region of interest.

The first ROI setting unit sets the left-most solid line of the reverse projection converted image as a reference lane, determines a dotted line located within a predetermined reference distance from the reference lane as a lane, and located within the reference distance from the lane. One solid line may be set as the edge lane, and a region from the reference lane to the edge lane may be detected as the first ROI.

The road area estimator may be configured to map the second region of interest to the first region of interest so that the common region of the first region of interest and the second region of interest has a polygonal shape based on the reference lane and the edge lane. The region of interest may be extracted to determine the final region of interest.

According to one aspect of the present invention, the position of the lane can be accurately estimated by semantically distinguishing the road image even in a situation where visibility of the lane included in the road image is significantly low, such as night driving, rainy weather, and lane loss problems. By actively setting the ROI, the lane detection rate and the processing speed required for lane detection can be improved.

1 and 2 are block diagrams showing a schematic configuration of a lane detection apparatus according to an embodiment of the present invention.
3 and 4 illustrate embodiments of setting a first region of interest and a second region of interest.
5 is a diagram illustrating an example of setting a final region of interest.
6 and 7 are flowcharts illustrating a schematic flow of a lane detection method according to an embodiment of the present invention.

DETAILED DESCRIPTION The following detailed description of the invention refers to the accompanying drawings that show, by way of illustration, specific embodiments in which the invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. It should be understood that the various embodiments of the present invention are different but need not be mutually exclusive. For example, certain shapes, structures, and characteristics described herein may be embodied in other embodiments without departing from the spirit and scope of the invention in connection with one embodiment. In addition, it is to be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the invention. The following detailed description, therefore, is not to be taken in a limiting sense, and the scope of the present invention, if properly described, is defined only by the appended claims, along with the full range of equivalents to which such claims are entitled. Like reference numerals in the drawings refer to the same or similar functions throughout the several aspects.

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

1 is a block diagram showing a schematic configuration of a lane detection apparatus according to an embodiment of the present invention.

Specifically, the lane detection apparatus 1 according to an embodiment of the present invention includes a preprocessing unit 10, a driving environment determination unit 20, a road area estimation unit 30, and a driving lane detection unit 40.

The preprocessor 10 may perform a preprocessing process for emphasizing the lane area in the input road image. The road image may be obtained from a photographing apparatus mounted on or installed in a vehicle, and the photographing apparatus may be any one of various devices mounted on the vehicle such as a camera module of a black box and photographing the front or rear of the vehicle to generate a road image. have.

As part of the preprocessing process, the preprocessor 10 may convert the road image into a gray scale image or a binary image. In this process, the preprocessor 10 may remove the influence of the camera internal parameters so that lens distortion by the imaging device does not occur. In addition, since the road image photographed by the photographing apparatus may include a partial region of the vehicle, the preprocessing unit 10 may cut a portion of the lower end of the distortion corrected image to remove information other than the lane in some cases. Can be.

Thereafter, the preprocessor 10 may generate a top-hat converted image in which a lane area is emphasized while noise is removed from the input road image by using a top-hat filter. However, the preprocessing unit 10 does not necessarily need to convert the image to the top-hat filter. You can also preprocess the road image. Hereinafter, for convenience of description, the road image in which the preprocessing process is performed by the preprocessor 10 will be defined and described as a preprocessed road image.

The driving environment determiner 20 may determine whether the lane is easily detected from the input road image by analyzing a road image to determine a situation of a road on which the vehicle is currently driving. For example, in rainy conditions, blur may occur in the road image photographed due to rain or cloud, which may cause difficulty in detecting a lane compared to the road image photographed on a clear day.

To this end, the driving environment determiner 20 may extract the brightness value by converting the RGB value of the road image into a YCbCr value. This is expressed as an equation.

In the YCbCr value, the Y value corresponds to the brightness of the road image, and thus, the driving environment determiner 20 may extract the brightness value of the road image by using the above-described equation.

If the calculated brightness value is greater than or equal to a predetermined threshold value, the driving environment determiner 20 determines that the contrast ratio between the road surface and the lane is clear and makes it easy to detect the lane from the road image. If the brightness value is less than a predetermined threshold value, it may be determined that the road image is difficult to detect lanes because visibility of the road image is poor. The driving environment determiner 20 may provide the determined result to the road area estimator 30.

The road area estimator 30 may estimate the road area from the road image by using the preprocessed road image and set the area of interest. In this case, the road area estimation unit 30 may set the ROI in different ways according to the result determined by the driving environment determination unit 20. To this end, as shown in FIG. 2 to be described later, the road area estimator 30 may include a first ROI setting unit 31 and a second ROI setting unit 32.

The road area estimator 30 sets the final ROI by using the first ROI set by the first ROI setter 31 or the second ROI set by the second ROI setter 32. Can be. In this case, as described above, the road area estimator 30 may set the ROIs in different ways according to the result determined by the driving environment determiner 20.

In detail, when it is determined that the brightness value of the road image is greater than or equal to the threshold value, the road region estimator 30 may set the first ROI as the final ROI. That is, when the road region estimator 30 determines that the road image is sufficiently bright to distinguish lanes only by the image processing technique performed by the first ROI setting unit 31, the first ROI setting unit 31 The region of interest set by the user may be set as the final region of interest (road region).

On the other hand, when it is determined that the brightness value of the road image is less than the threshold value, the road region estimator 30 controls the second ROI to be additionally set through the second ROI setting unit 32 and controls the first ROI. The final region of interest may be set in consideration of the second region of interest. Detailed description thereof will be described later.

The driving lane detector 40 may detect at least one driving lane from the final ROI.

The driving lane detection unit 40 may detect at least one driving lane from the final region of interest using a Hough transform and a random SAmple consensus (RANSAC) algorithm. First, the driving lane detection unit 40 may calculate respective linear equations for the points constituting the straight lines included in the final ROI by using the Hough transform, and extract intersection points of the straight lines generated according to the straight line equations. If the number of straight lines passing through any one intersection is equal to or greater than a predetermined number, the driving lane detection unit 40 determines that the points corresponding to the straight lines passing through the intersection are points forming a straight line, and the driving lanes are determined as these. Can be set as a candidate group. The driving lane detection unit 40 extracts two arbitrary points, connects the two points in a straight line, performs the same process for all the points, and then uses the RANSAC algorithm to finally determine the driving lane candidate group receiving the most support. It can be detected by the traveling lane.

The Hough transform is a technique for detecting linear components in an image, and the RANSAC algorithm extracts a model with the most agreement among inlier information. The HANS transform and RANSAC algorithm are used for image processing technology and regression analysis. Since the technology is well known in the description of each specific algorithm will be omitted.

In this case, since the final ROI is set as an area estimated as the road area, the driving lane detector 40 may detect the lane candidate group and the driving lane in consideration of only the straight lines of the final ROI, not all the straight lines of the road image. Rate and throughput can be improved.

FIG. 2 is a diagram illustrating a specific configuration of the road area estimating unit 30 of FIG. 1.

In detail, the road area estimator 30 according to an embodiment of the present invention may include a first ROI setting unit 31 and a second ROI setting unit 32.

The first ROI setting unit 31 may set a first ROI estimated as a road region from the preprocessed road image. In this case, the first ROI setting unit 31 may set the first ROI from the reverse projection converted image of the preprocessed road image. In this regard, it will be described with reference to FIG.

3 illustrates an example in which the first ROI setting unit 31 sets the first ROI. FIG. 2A is a diagram illustrating an example of an original image of a road image, and FIG. 2B is a diagram illustrating an example of a reverse projection converted image.

As described above, the first ROI setting unit 31 may set the first ROI from the reverse projection converted image. The first ROI setting unit 31 may generate a reverse projection converted image obtained by converting a road image as illustrated in FIG. 2A into an image from which perspective is removed as illustrated in FIG. 2B.

In order to estimate the road width, it is necessary to detect the boundary between the road and the non-road. The road boundary is structurally different from other lanes. In Korea, since the vehicle travels to the right side, the solid line on the left side may be the center line. In this case, the solid line at the right edge is often the end of the lane, and thus the road width can be estimated from the center line and the solid line at the right edge. In general, the widths of the driving lanes are spaced apart by a predetermined interval.

Using this feature, the first ROI setting unit 31 may detect the solid line located to the left of the reverse projection converted image and set it as the reference lane. The first ROI setting unit 31 may determine a dotted line located within a preset reference distance in a right direction with respect to the reference lane as the lane. If there is a lane determined as a dotted line, the first ROI setting unit 31 may check whether there is a dotted line (second lane) located within a reference distance in the right direction again based on the determined dotted line (first lane). have. Here, the reference distance may be set to 3m to 3.5m, but the reference distance is not limited thereto and may be set to various values according to road conditions for each country or region.

In this way, the first ROI setting unit 31 may detect the lanes arranged in the road area by repeatedly performing the above-described method until the solid line is detected in the right direction. Thereafter, the first ROI setting unit 31 may estimate the area from the set reference lane to the edge lane as the road area, and set it as the first ROI (shaded area of the drawing).

In some other embodiments, the first ROI setting unit 31 may set the first ROI in another method according to the brightness value of the reverse projection converted image.

For example, in a tunnel, all lanes are arranged in solid lines to prohibit lane movement of the vehicle. In this case, when setting the first ROI using the above-described method, a problem of estimating only one lane as a road area may occur. Therefore, it is necessary to extract the region of interest in different ways depending on whether the road is outdoors or inside the tunnel.

To this end, the first ROI setting unit 31 extracts the brightness value of the reverse projection converted image, and if it is determined that the brightness value of the reverse projection converted image is smaller than a predetermined reference value, it is determined that the vehicle is currently located inside the tunnel. In another way, the first ROI may be set. In general, since the road image photographed inside the tunnel has a darker brightness value than the road image photographed in rainy weather, the reference value is preferably set to a value smaller than the above-described threshold value.

In detail, when the brightness value of the reverse projection conversion image is smaller than the reference value, the first ROI setting unit 31 may further include a solid line located within a reference distance in a right direction from the reference lane set as the rightmost solid line of the reverse projection conversion image. It can be judged that The first ROI setting unit 31 may repeat the process until the other solid line located within the reference distance is not detected based on the determined lane to distinguish the lanes of the road. In this case, when the solid line is detected within a preset proximity distance, the first ROI setting unit 31 may determine that the first ROI is not a normal lane. For example, a shoulder in a tunnel may be set to be narrower than a normal lane width, and a shoulder is a lane used in an emergency situation and needs to be excluded from the road area.

Therefore, when the lane of interest is detected within a close distance based on the last detected lane, or the lane is not detected within the reference distance, the first ROI setting unit 31 edges the last detected lane. Can be set. In this case, the proximity distance may be set to be smaller than the reference distance. For example, the reference distance may be set to a value between 3m and 3.5m, and the proximity distance may be set to 2.5m, but the proximity distance is smaller than the reference distance. When set, no specific numerical value is imposed. The first ROI setting unit 31 may estimate a region from the reference lane to the edge lane as the road region and set it as the first ROI.

The second region of interest setting unit 32 may set a second region of interest estimated from the road image as a road region by using the previously learned deep neural network. In this regard, it will be described with reference to FIG.

4 is a diagram illustrating an example in which a second ROI is set by the second ROI setting unit 32.

The second ROI setting unit 32 may perform an encoder process and a decoder process of the road image by using the deeply trained deep neural network.

In the encoder process, the second ROI setting unit 32 may extract a feature map by applying a convolution filter to the road image. The second ROI setting unit 32 may reduce the size of the feature map by performing a maximum plooing process of the generated feature map. The second ROI setting unit 32 repeats the convolutional layer and the maximum pooling layer a predetermined number of times, and then, in the decoder process, up-samples the maximum pooled RGB values through a deconvolution process, thereby reconstructing the road image. Can be restored to the size of. As shown in (b) of FIG. 4, the road image that has undergone this process may be converted into an image having the same color among pixels having similar characteristics.

That is, the second ROI setting unit 32 may label the pixels having similar characteristics among the pixels constituting the road image by using the deep neural network to have the same color. Subsequently, the second ROI setting unit 32 may search for pre-learned class information to distinguish which objects are pixels labeled with the same group. Accordingly, the second region of interest setting unit 32 may estimate the object region labeled with a specific color (in the illustrated embodiment, the purple region) as the road region and set it as the second region of interest.

Here, the pre-learned class information may include classes for objects that can be classified in the road image, such as roads, lanes, fences, buildings, and the like. In addition, the pre-learned deep neural network may be in the form of a SegNet neural network algorithm, which is a kind of convolutional neural network.

Thereafter, the second ROI setting unit 32 may update the parameters constituting the model through a backpropagation process of the extracted probability result. Finally, the learned parameters can be reconstructed into RGB images for semantic classification.

As described above, when it is determined that the brightness value of the road image is less than the threshold value, the road region estimator 30 may set the final ROI by considering the first ROI and the second ROI. In this regard, it will be described with reference to FIG.

5 is a diagram illustrating an example in which the road area estimation unit 30 sets a final ROI.

In order to prevent a problem in which the first ROI cannot be determined when the road image such as the reference lane or the edge lane cannot be detected by the first ROI setting unit 31 due to poor quality of the road image, If it is determined that the visibility of the road image is not good, the estimator 30 may perform the operation of merging the second ROI set by the second ROI setting unit 32 with the first ROI.

For example, the road region estimator 30 may set the final ROI by mapping the second ROI to the first ROI as illustrated. The road region estimator 30 may map the second region of interest to the first region of interest and set a common region of the first region of interest and the second region of interest as the final region of interest. In this case, the final ROI set by the road area estimator 30 may be set as a polygonal area based on the reference lane and the edge lane.

In detail, the road area estimator 30 may divide the second ROI by lane. As described above, the second ROI setting unit 32 may label pixels constituting the road image with pixels having a specific meaning. In this case, the type of the object labeled by the second region of interest setting unit 32 may include a lane, a road, a fence, a vehicle, and an uninterested object. The second ROI setting unit 32 may set an area including the lane and the road object as the second ROI, and accordingly, the road area estimation unit 30 separates the second ROI by lanes. Primary, secondary, and tertiary lanes). The road area estimator 30 may extract the height values h1, h2, and h3 for each lane.

Finally, the road area estimator 30 sets the width of the final ROI from the reference lane set from the first ROI to the edge lane, analyzes the second ROI, and divides the height value of each lane divided by the lanes. By connecting (h1, h2, h3) with the height value for each lane adjacent to each other and setting the height value for each section of the polygonal shape, it is possible to set the height value of the final ROI by car. Accordingly, the road area estimator 30 determines and removes a straight line detected outside the area estimated as the road area as noise, and estimates the straight line or the dotted line detected in the area estimated as the road area as a lane. The final ROI may be set in the image.

As another example, the road area estimator 30 may calculate a similarity degree between the first ROI and the second ROI and set the final ROI based on the similarity.

The road area estimator 30 may set the final ROI that is a result of the merged finally, and the driving lane detector 40 may extract the final road area based on this.

6 and 7 are flowcharts illustrating a schematic flow of a lane detection method according to an embodiment of the present invention.

First, as shown in FIG. 6, when the lane detecting apparatus 1 according to the present invention receives a road image for the front of the vehicle from a photographing apparatus mounted on the vehicle, the lane detecting apparatus 1 may extract a brightness value of the road image (510). ). In this case, the lane detection apparatus 1 according to the present invention may set the final ROI in different ways according to a result of comparing the brightness value with the threshold value.

Specifically, when it is determined that the extracted brightness value is greater than or equal to a preset threshold value (YES at 520), the lane detection apparatus 1 sets the first ROI set through the first ROI setting unit 31 as the final ROI. Can be set (530). On the other hand, if it is determined that the extracted brightness value is less than the preset threshold value (NO in 520), the lane detection apparatus 1 may set the final ROI after performing the subsequent process 535. In this regard, it will be described with reference to FIG.

7 is a flowchart illustrating a specific flow of the subsequent processing 535 of the lane detection apparatus 1.

When it is determined that the brightness value is less than the preset threshold value, the line detection device 1 may be provided with the first region of interest set by the first region of interest setting unit 31 (535_1). In detail, the first ROI setting unit 31 of the lane detection apparatus 1 is a kind of image processing technology. The first ROI setting unit 31 may generate the first ROI based on the reference lane and the reference distance from the reverse projection converted image.

At the same time, the lane detection apparatus 1 may be provided with the second region of interest set by the second region of interest setting unit 32 (535_2). The second ROI setting unit 32 of the lane detection apparatus 1 may semantically classify the road image by using a deep neural network method. Specifically, the pixel group labeled by labeling the pixels of the road image may be used. You can tell if it's a kind of object. The second ROI setting unit 32 may set the second ROI based on the pixel group divided into the road object.

The lane detection apparatus 1 may set the final ROI in consideration of the first ROI and the second ROI (535_3).

Thereafter, the lane detection apparatus 1 may extract the driving lane candidate group from the set final region of interest (540). Finally, the lane detecting apparatus 1 may detect at least one driving lane from the extracted driving lane candidate group (550).

Such a technique for providing a lane detection method may be implemented in the form of program instructions that may be implemented as an application or executed through various computer components, and recorded on a computer readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, etc. alone or in combination.

The program instructions recorded on the computer-readable recording medium are those specially designed and configured for the present invention, and may be known and available to those skilled in the computer software arts.

Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical recording media such as CD-ROMs, DVDs, and magneto-optical media such as floptical disks. media), and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like.

Examples of program instructions include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules to perform the process according to the invention, and vice versa.

Although the above has been described with reference to the embodiments, those skilled in the art will understand that various modifications and changes can be made without departing from the spirit and scope of the invention as set forth in the claims below. Could be.

1: lane detection device
10: pretreatment unit
20: driving environment determination unit
30: road area estimation unit
31: first region of interest setting unit
32: second region of interest setting unit
40: driving lane detection unit

Claims (15)

Extracting a brightness value of a road image and comparing the brightness value with a preset threshold value;
Setting a first ROI estimated as a road area from a reverse projection transformation image of the road image;
If it is determined that the brightness value is smaller than the threshold value, setting a second ROI including an object determined to be a lane and a road from the road image by using a pre-learned deep neural network;
Determining a final region of interest estimated as a road region in the road image based on the first region of interest and the second region of interest; And
Detecting at least one traveling lane from the last region of interest using a predetermined lane detection algorithm.
The method of claim 1,
Setting to the second region of interest,
Each pixel constituting the road image is labeled with any one of a pre-learned class by using the deep neural network, and the pixels of the road image are grouped into objects for the learned class and grouped into a road area class. Setting the selected pixels as the region of interest.
The method of claim 1,
Setting as the first region of interest,
The leftmost solid line of the reverse projection converted image is set as a reference lane, a dotted line located within a predetermined reference distance from the reference lane is determined as a lane, and any one solid line located within the reference distance from the lane is the edge lane And detecting the area from the reference lane to the edge lane as the first region of interest.
The method of claim 3,
Setting as the first region of interest,
If the brightness value of the reverse projection converted image is smaller than a reference value, the solid line located within a predetermined reference distance from the reference lane is further determined as a lane, and the rightmost lane among the lanes continuously determined based on the reference distance is the edge. The lane detection method, and the reference value is set to a value smaller than the threshold value, lane detection method.
The method of claim 3,
Determining the final region of interest is
The second region of interest is mapped to the first region of interest, and a common region of the first region of interest and the second region of interest is extracted as a region having a polygonal shape based on the reference lane and the edge lane. The lane detection method of determining to the region of interest.
The method of claim 6,
Extracting to the area of the polygonal form,
And setting the height value of each section of the polygonal area using the height of each lane divided by the lane from the reference lane to the edge lane and setting the width of the final ROI. Detection method.
The method of claim 1,
Detecting the at least one driving lane,
Convert any one point constituting a straight line of the final region of interest to a straight line equation, detect an intersection where a plurality of straight line equations converted for each point cross each other, and the number of straight line equations passing through the intersection is critical And extracting a driving lane candidate group using linear equations having a value or more.
The method of claim 1,
Detecting the at least one driving lane,
And detecting the at least one driving lane from the driving lane candidate group using a RANSAC algorithm.
The method of claim 1,
Determining the final region of interest is
And if it is determined that the brightness value is smaller than the threshold value, determining the detected first ROI as the final ROI.
The method of claim 1,
Before extracting the brightness value of the road image, further comprising the step of performing a pre-processing of the road image,
Performing the pretreatment process,
And converting the road image into a gray scale image, and converting the gray scale image using a top-hat filter.
A computer-readable recording medium having a computer program recorded thereon for performing the lane detection method according to any one of claims 1 to 10.
A driving environment determination unit which extracts a brightness value of a road image and compares the brightness value with a preset threshold value;
A first ROI setting unit configured to set a first ROI estimated as a road region from an inverse projection transform image of the road image, and when the brightness value is smaller than the threshold value, a deeply trained deep neural network is used. And a second ROI setting unit configured to set a second ROI including an object determined to be a lane and a road from the road image, and based on the first ROI and the second ROI, a road region in the road image. A road area estimator for determining a final region of interest estimated to be; And
And a lane detector for detecting at least one driving lane from the last region of interest using a predetermined lane detection algorithm.
The method of claim 12,
The second region of interest setting unit,
Each pixel constituting the road image is labeled with any one of a pre-learned class by using the deep neural network, and the pixels of the road image are grouped into objects for the learned class and grouped into a road area class. And set the pixels as the second region of interest.
The method of claim 12,
The first ROI setting unit may include
The leftmost solid line of the reverse projection converted image is set as a reference lane, a dotted line located within a predetermined reference distance from the reference lane is determined as a lane, and any one solid line located within the reference distance from the lane is the edge lane And detecting the area from the reference lane to the edge lane as the first region of interest.
The method of claim 14,
The road area estimation unit,
The second region of interest is mapped to the first region of interest, and a common region of the first region of interest and the second region of interest is extracted as a region having a polygonal shape based on the reference lane and the edge lane. A lane detecting apparatus for determining the region of interest.

Images