KR20130000023A - Method for dectecting front vehicle using scene information of image - Google Patents

Abstract

PURPOSE: A front vehicle detecting method using scene information of an image is provided to improve a front vehicle detection speed and detecting performance by utilizing the scene information of a relatively simple image in detecting a vehicle candidate area and indentifying a vehicle. CONSTITUTION: A driving assistance system receives an image obtained from the camera of a running vehicle(S100). The system preprocesses the input image(S200). The system determines a ROI(Region Of Interest) for a vehicle candidate area extraction(S300). The system extracts a road area and a shadow area(S400). The system extracts a vehicle candidate area(S500). The system verifies the presence of a vehicle in the extracted vehicle candidate area(S600). [Reference numerals] (AA) Start; (BB) End; (S100) Inputting an image; (S200) Preprocessing the image; (S300) Determining a region of interest for extracting a vehicle candidate area; (S400) Extracting a road area and a shadow area; (S500) Extracting the vehicle candidate area; (S600) Verifying the vehicle candidate area

Description

Method for detecting front vehicle using scene information of image {Method for dectecting front vehicle using scene information of image}

The present invention relates to a method for detecting a front vehicle using scene information of an image, and more particularly, to a method for detecting a front vehicle of a driving vehicle during the day using scene information of an image obtained from a camera mounted on the driving vehicle. will be.

Research and development on the acquisition and utilization of driving information is active to assist the driver's driving for the prevention of driving accidents, and many driver driving assistance systems are already being used commercially in high-grade passenger cars. Although various sensors are used to obtain driving information, research on image processing based driver assistance system is actively conducted because image information acquired through a camera provides more information than driving information obtained from other sensors.

One of the pieces of information essential to the development of driver assistance systems is forward vehicle detection. Front vehicle detection is generally performed in two stages: 'Hypothesis Generation' and 'Hypothesis Verification' to confirm whether the candidate region includes a vehicle.

According to the prior art, the existing front vehicle candidate area detection includes symmetry, color, shadow, corner, vertical and horizontal edge, texture, vehicle traffic light, etc., and the existing vehicle verification means template matching, vehicle characteristic information. Based forward vehicle classifier training and verification, tracking information based matching, etc. are used. Template matching as a verification means does not have high accuracy, and feature information-based classifier training and verification has a problem in that it is not suitable for real-time execution because the offline training is complicated and the online verification algorithm is complicated in some cases.

In addition to the above-described prior art, in the case of other conventional techniques, it takes a lot of time because the entire image data is searched for extracting the vehicle candidate region, and a region where a vehicle cannot exist (the vehicle candidate region due to an off-road region, a shadow, etc.). In the case of a road region which is mistaken for a road, etc.), a vehicle candidate region is extracted, which causes a problem of low vehicle detection accuracy.

SUMMARY OF THE INVENTION The present invention has been made to overcome the above-mentioned conventional problems, and the problem to be solved by the present invention is the speed and detection of forward vehicle detection by utilizing scene information of an image, which does not require a large amount of calculation, for vehicle candidate region detection and vehicle verification. An object of the present invention is to provide a method for detecting a front vehicle using scene information of an image capable of improving performance.

According to an exemplary embodiment of the present invention, the method includes: inputting an image acquired by a camera mounted on a driving vehicle; Performing image preprocessing to remove noise of an input image and to improve edge information; Determining a region of interest for vehicle candidate region extraction; Extracting road and shadow areas by extracting road and shadow information; Detecting horizontal and vertical lines of the vehicle and extracting a vehicle candidate region including the detected horizontal and vertical lines; And verifying whether a vehicle exists in the extracted vehicle candidate region, a front vehicle detection method using scene information of an image.

The performing of the image preprocessing may include performing gray level image conversion when the input image is color; Filtering for noise improvement and edge information improvement; And detecting a horizontal edge and a vertical edge using a Sobel edge detector.

The filtering may be performed by using a rank order filter.

The determining a region of interest may include estimating a vanishing point; Obtaining a horizontal line using the estimated vanishing point; And determining a region of interest for vehicle candidate region extraction, wherein the region of interest is determined as a space between a portion below the line 30 pixels higher than the horizontal line and 1/8 or more of the height of the entire image data. .

The estimating the vanishing point may include performing an image preprocessing process for estimating the vanishing point; A huff line identification step of identifying a huff line by applying a hough transform on edges obtained in the image preprocessing, and then integrating similar huff lines having similar directions and adjacent positions; And estimating a vanishing point from the candidate hough line.

The extracting the road area and the shadow area may include determining a road statistical information ROI for extracting the road area and the shadow area; Extracting road statistics information from the determined road statistical information ROI, comprising: obtaining an average m and a standard deviation v of gray level pixel values of the image image of the ROI; And extracting a road area and a shadow area based on the road statistical information.

The extracting of the road area and the shadow area may be performed if any pixel gray level value L is in the range of Equation 1 below.

[Formula 1]

Mean (m) -3 * standard deviation (v) <L <mean (m) + 3 * standard deviation (v)

The pixel is regarded as a road candidate pixel, and if the number of road candidate pixels in any area exceeds 60%, the area is extracted as a road area.

The extracting of the road area and the shadow area may be performed if any pixel gray level value L satisfies Equation 2 below.

[Formula 2]

L <mean (m)-4 * standard deviation (v)

It is regarded as a shadow candidate pixel, and if a shadow candidate pixel exceeds 50% in an arbitrary region, the corresponding region is extracted as a shadow region.

The extracting of the vehicle candidate region may include detecting horizontal and vertical lines from image data in which image preprocessing is performed; And extracting a vehicle candidate region by using the detected horizontal and vertical lines, wherein a horizontal line exists and includes a horizontal line and a vertical line when a vertical line is detected near the lower portion or the upper portion of the horizontal line. Extracting the rectangle as the vehicle candidate region.

The detecting of the horizontal and vertical lines may be performed by applying a Hough transform on an edge binarized image obtained by binarizing an edge image obtained by applying a Sobel edge mask.

The vehicle candidate region verification step may include: checking whether a shadow region exists in the vehicle candidate region; Checking whether a road area exists at a lower portion of the vehicle candidate area; Identifying additional verification conditions; And verifying whether the vehicle exists in the vehicle candidate region according to the verification weight sum by applying a weight to each verification result.

The additional verification condition may be any one or a combination of two or more of the presence or absence of wheels, symmetry, horizontal edge distribution, and tail light in the vehicle candidate region.

In the verifying whether the vehicle exists in the vehicle candidate region according to the verification weight sum, the verification weight sum score is set higher as the vehicle width is wider to verify whether the vehicle exists.

As in the present invention, vehicle candidate region detection and vehicle verification of scene information of a relatively simple image that does not require a large amount of computation such as scene geometric information such as vanishing point and horizon of the image scene and scene appearance information such as road area gray level information. The speed of the front vehicle detection is improved and the detection performance is improved, and the front vehicle detection method proposed by the present invention enables the development of an improved image processing based real-time driver assistance system.

1 is a schematic flowchart of a method for detecting a front vehicle using scene information of an image according to an exemplary embodiment of the present invention.
FIG. 2 is a flowchart illustrating a process for performing the image preprocessing step shown in FIG. 1.
3 is an image photograph to which the image preprocessing step is applied.
4A is a horizontal edge mask used for horizontal edge detection in this embodiment, and FIG. 4B is a vertical edge mask used for vertical edge detection.
5A to 5D show an image obtained by applying an edge mask and binarized and an image obtained by applying an edge mask after applying a filter and binarized.
6 is a flowchart illustrating a process for performing a region of interest determination.
7 is a diagram illustrating a photograph in which a region of interest is determined.
8 is a flowchart illustrating a detailed process for performing the vanishing point estimation process illustrated in FIG. 6.
9 is a flowchart illustrating a process for performing a road and shadow information extraction step.
FIG. 10A illustrates a state in which a road statistical information ROI is determined, FIG. 10B illustrates a state in which a road region is determined, and FIG. 10C illustrates a state in which a shadow region is determined.
11 is a flowchart illustrating a process for performing a vehicle candidate region extraction step.
12 is a diagram illustrating a horizontal line, and FIG. 13 is a diagram illustrating a vertical line.
14 is a diagram illustrating a vehicle candidate area including a horizontal line and a vertical line.
15 is a flowchart illustrating a process for performing a vehicle candidate region verification step.
16 shows the resultant image when the shadow area exists.
17 shows the resultant image when the road area exists.
18 is an image showing a front vehicle detected by using a front vehicle detection method using scene information of an image according to an exemplary embodiment of the present invention.

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

1 is a schematic flowchart of a method for detecting a front vehicle using scene information of an image according to an exemplary embodiment of the present invention.

Referring to FIG. 1, first, a process of inputting an image acquired by a camera mounted on a driving vehicle is performed (S100).

An image preprocessing step for removing noise of an input image and improving edge information is performed (S200).

Then, a step of determining a region of interest for vehicle candidate region extraction is performed (S300). In order to extract candidate regions, horizontal and vertical edge detection methods are used in the image data. In this case, since it takes a long time to detect the horizontal edge and the vertical edge in all regions of the image data, the embodiment of the present invention determines the region of interest and extracts the vehicle candidate region within the region of interest.

In front vehicle detection, the area of interest is effectively a road area. However, since it is difficult and time consuming to accurately extract only the road area from the image data, in the embodiment of the present invention, a horizontal line is obtained by using vanishing points, and the region of interest is based on the horizontal line. Decide Determining the region of interest is described in more detail below.

Then, the road and shadow information is extracted from the image data to extract the road area and the shadow area (S400).

Detecting horizontal and vertical lines of the vehicle, and extracting a vehicle candidate region including the detected horizontal and vertical lines (S500).

Then, a step of verifying whether a vehicle exists in the extracted vehicle candidate region is performed (S600).

2 is a flowchart illustrating a process for performing the image preprocessing step shown in FIG. 1, FIG. 3 is an image photograph to which the image preprocessing step is applied, and FIG. 4A is a horizontal edge mask used for horizontal edge detection in the present embodiment. 4B is a vertical edge mask used for vertical edge detection, and FIGS. 5A to 5D show an image obtained by applying an edge mask and binarizing the edge mask, and an image applying the edge mask after applying the filter and binarization.

Referring to FIG. 2, the image preprocessing step (S200) will be described. If the input image is color, the gray level image conversion process is performed (S210). Next, a filtering process for noise improvement and edge information improvement is performed (S220). Then, the horizontal edge and vertical edge detection process is performed using the Sobel edge detector (S230).

The color input image is disadvantageous in real time processing because it takes more time to process the image than the gray level image. In addition, in the case of color image data, since a lot of color information disappears at night, gray level based image processing is preferable for the same algorithm processing day and night.

Since the gray level image obtained from the input image is noisy and the edges are not clear in some cases, the filtering process is performed. In an embodiment of the present invention, a rank order filter was implemented and used. 3A illustrates a color image, and FIG. 3B illustrates a gray level image converted through S210, and FIGS. 3C and 3D show an image to which a horizontal rank order filter and a vertical rank order filter are applied.

In the exemplary embodiment of the present invention, a Sobel edge detector having a small amount of calculation is used for a real-time processing rather than a large calculation canny edge detector. Sobel horizontal edge masks and vertical edge masks used for horizontal edge detection are shown in FIGS. 4A and 4B, respectively.

FIG. 5A is a binarized image obtained by applying a horizontal Sobel edge mask, and FIG. 5B is binarized image after applying a horizontal Sobel edge mask after applying a horizontal rank order filter. In addition, FIG. 5C is a binarized image obtained by applying a vertical Sobel edge mask, and FIG. 5D is binarized image after applying a vertical Sobel edge mask after applying a vertical rank order filter.

As can be seen in the comparison picture of FIG. 5, after applying the rank order filter, it can be seen that the Sobel horizontal and vertical edge detections have more noise and sharper edges.

FIG. 6 is a flowchart illustrating a process for performing a region of interest determination, FIG. 7 is a diagram illustrating a photograph in which a region of interest is determined, and FIG. 8 is a flowchart illustrating a detailed process for performing the vanishing point estimation process illustrated in FIG. 6. to be.

Referring to FIGS. 6 to 8, a process of determining a region of interest (S300) is performed. First, a process of estimating a vanishing point is performed (S310).

The process of obtaining a horizontal line using the vanishing point estimated through the process S310 is performed (S320).

Thereafter, a process of determining a region of interest for vehicle candidate region extraction is performed (S330). In most road images, vehicles appear above 1/8 of the image data height, and the vehicle is visible over the horizon. Therefore, according to the exemplary embodiment of the present invention, as shown in FIG. 7, the space between the portion below the line 30 pixels higher than the horizontal line and 1/8 or more of the height of the entire image data is determined as the region of interest as the region of interest for vehicle candidate region extraction. It was. Hereinafter, horizontal and vertical edges are extracted only within the region of interest, and vehicle candidate regions are extracted using the same.

Referring to Figure 8 looks at the process of estimating the material point in more detail.

Vanishing point refers to a 2D image point where 3D real world parallel lines are projected and intersected on a 2D image. Since the lanes of 3D real-world roads are parallel, vanishing points that appear as a point in the road driving 2D image can give lane information and are very useful for driver assistance systems. In the embodiment of the present invention is used as a prior material for obtaining a horizontal line.

In order to estimate the vanishing point, first, an image preprocessing process for estimating the vanishing point is performed (S311). The preprocessing process is a process of performing image preprocessing so that the edge portion from the lane is more clearly extracted by clarifying the lane portion in the road driving image data. First, a color image is converted into a gray level image for the purpose of real-time processing and the same day and night image processing. Next, the area of the road area is calculated, and the portion of the road area that is consistent with the road area statistics is regarded as the road area, and the brightness intensity is assigned to 0. In addition, a rank order filter is applied so that the edges appear more clearly. After applying the edge detector and then binarized to obtain a binarized edge image.

Then, the Huff line identification process is performed (S312). The huff line identification process is a process of obtaining candidate huff lines necessary for the vanishing point prediction for the edges obtained in the preprocessing process S311. After applying the Hough transform on the edges to find the Hough line, the similar Huff lines with similar directions and very close positions are identified to identify candidate Hough lines to be used for the vanishing point estimation.

Then, a process of estimating the vanishing point is performed (S313). It is a process of estimating a vanishing point from a candidate huff line. In order to calculate the reliability of the estimated vanishing point and improve the reliability, the vanishing point estimate is improved by using the past vanishing point estimate.

FIG. 9 is a flowchart illustrating a process for performing a road and shadow information extraction step, FIG. 10A is a diagram illustrating a state of interest of road statistical information, and FIG. 10B is a diagram of a state of determining a road area, and FIG. 10C. Is a view showing a state in which the shadow area is determined.

Referring to FIG. 9, the road and shadow information extraction step is performed. First, a process of determining a road statistical information ROI for extracting a road area and a shadow area is performed (S410). First, in order to determine the ROI of the statistical information of the road gray level, the road statistical information ROI, which is mainly included in the road information, is vertical from the 2/3 of the height of the image image based on the upper end of the image image. Up to 5/6 of the height, the area from the 1/3 to 2/3 point of the left side of the video image is defined as the road statistical information region of interest. The reason why this area is regarded as the road statistical information region of interest is that this part of the road driving image image is usually a road area without a vehicle being detected (see FIG. 10A).

Next, a process of extracting road statistics information from the determined road statistics information ROI is performed (S420). In the present embodiment, the mean (m) and standard deviation (v) of gray level pixel values of the video image of the ROI are obtained.

Then, a process of extracting the road area and the shadow area is performed (S430). In an embodiment of the present invention, if any pixel gray level value L is in the range of [Equation 1] below, the pixel is regarded as a road candidate pixel, and if the number of road candidate pixels in any area exceeds 60%, the corresponding area Is regarded as the road area.

[Formula 1]

Mean (m) -3 * standard deviation (v) <L <mean (m) + 3 * standard deviation (v)

FIG. 10B shows a binarized image of the road candidate pixel area with the road area candidate pixels white and other pixels black for the image of FIG. 10A.

In addition, when an arbitrary pixel gray level value L satisfies the following [Equation 2], it is regarded as a shadow candidate pixel, and when a shadow candidate pixel exceeds 50% in any region, the corresponding area is regarded as a shadow area.

[Formula 2]

L <mean (m)-4 * standard deviation (v)

FIG. 10C shows a binarized image of the shadow candidate pixel area in which the shadow area candidate pixels are white and other pixels are black for the image of FIG. 10A.

11 is a flowchart illustrating a process for performing a vehicle candidate region extraction step, FIG. 12 is a diagram illustrating a horizontal line, FIG. 13 is a diagram illustrating a vertical line, and FIG. 14 includes a horizontal line and a vertical line. It is a figure which detected the vehicle candidate area | region to make.

Referring to FIG. 11, a vehicle candidate region extraction step (S500) is performed. First, a process of detecting horizontal and vertical lines from image data in which a preprocessing process is performed is performed (S510). Different views, such as the front vehicle view or the rear vehicle view of the vehicle, include many horizontal and vertical structures, such as the rear car's glass, bumpers, and the horizontal and vertical edges detected from these structures are powerful clues to the vehicle's presence. Horizontal and vertical edges detected outside the region of interest are excluded from the subject, even horizontal edges and vertical edges that do not have a certain length.

In the embodiment of the present invention, since the road is assumed to be flat in the driving vehicle image, only horizontal lines (that is, line segments parallel to the horizontal line) are considered in the horizontal edge image, and the vertical having a constant length in the vertical edge image is considered. Review only the lines.

In the embodiment of the present invention, a 3x3 Sobel edge mask with low processing time was used for edge detection, and a rank order filter was used before application of the Sobel edge detector to improve edge detection (image preprocessing step).

The horizontal and vertical lines are extracted by applying Hough transform on the edge binarized image obtained by binarizing the edge image obtained by applying the Sobel edge mask. In an embodiment of the present invention, lines are found by Hough transform, and horizontal lines and vertical lines are found among these lines by using angles formed by the horizontal lines.

FIG. 12 illustrates horizontal lines found by applying a Hough transform to the one-color video image, and FIG. 13 illustrates vertical lines found by applying the Hough transform to the one-color video image.

After detecting the horizontal and vertical lines through the process S510, the process of extracting the vehicle candidate region using the detected horizontal and vertical lines is performed (S520).

12 and 13, a vertical line may not be detected in a vehicle in which a horizontal line is detected, or a horizontal line may not be detected in a vehicle in which a vertical line is detected. In the exemplary embodiment of the present invention, when a horizontal line exists and a vertical line is detected near the lower portion of the horizontal line or near the upper portion, a quadrangle including the horizontal line and the vertical line is extracted as the vehicle candidate region.

FIG. 14 shows an example of the irradiation rectangle area around the selected horizontal line to check the vertical edge with respect to the detected horizontal line of FIG. 12.

FIG. 15 is a flowchart illustrating a process for performing a vehicle candidate region verification step, FIG. 16 illustrates a result image when a shadow region exists, FIG. 17 illustrates a result image when a road region exists, and FIG. 18. Is an image showing the front vehicle detected using the front vehicle detection method using the scene information of the image according to an embodiment of the present invention.

Road-driven vehicles vary in appearance in the image, depending on sunlight, distance, and viewing angle. In bright daylight vehicles, the sun's shadow appears under the car due to sunlight. Thus, shadows are a strong basis for vehicle identification. However, the shadows are clearly visible to the naked eye, which may not be apparent in the camera acquisition image or on the side rather than under the vehicle, and there are no shadows on cloudy days, in the rain, in the evening and at night.

Therefore, it is necessary to verify by considering several conditions rather than determining whether a vehicle exists by any one condition in the vehicle identification step. In the exemplary embodiment of the present invention, shadow information, road information, symmetry information, and wheel information of the vehicle are used for vehicle verification.

Looking at the vehicle candidate region verification step (S600) with reference to FIG. 15, first, a process of checking whether a shadow region exists is performed (S610).

The presence of the shadow area under the vehicle candidate area is a strong clue that the vehicle exists in the vehicle candidate area. The area to check the shadow checks whether there is a shadow at about 1/8 of the height of the candidacy above and below the horizontal line of the candidacy. FIG. 16 illustrates an image in a case where a shadow area exists at a lower portion of a vehicle candidate area.

Next, a process of checking a road area is performed (S620). It is as follows whether the road area exists in the lower part of the vehicle candidate area. In the embodiment of the present invention, it is checked whether there is a road area at a portion that is about 1/2 of the height of the vehicle candidate under the vehicle candidate. 17 shows an image of the result when the road area exists at the lower portion of the vehicle candidate area.

The process of checking the additional verification condition is performed (S630).

 The wheels of the vehicle are mostly black, with no change in color due to changes in sunlight or position. Therefore, it is important information that can be used for vehicle identification. Most vehicles maintain symmetry in the horizontal and horizontal directions. This information is used to distinguish between vehicles and non-vehicles. Typically, in vehicles, relatively many horizontal edges appear. However, the image obtained from the camera shows a lot of difference in the number of horizontal edges that can be checked according to the distance of the vehicle, and it depends on the color of the vehicle. Typically, in vehicles, relatively many horizontal edges appear. However, the image obtained from the camera shows a lot of difference in the number of horizontal edges that can be checked according to the distance of the vehicle, and it depends on the color of the vehicle.

Therefore, in the exemplary embodiment of the present invention, the presence of wheels, presence of symmetry, horizontal edge distribution, presence of tail lights, and the like are used as additional verification conditions.

Then, a process of verifying whether a vehicle exists in the vehicle candidate region is performed based on the weighted result by applying weights to the verification result (S640).

In the embodiment of the present invention, the verification weight according to the verification method is determined as follows.

Shadow Area Exists: +2 if true

-Road zone presence: +2 if true

-Both shadows and road zones exist: +4 if true

-Without both shadow and road area presence; -One

-Presence of symmetry; +2 if true

-Horizontal edge distribution; Max {(n-2), 5} for n horizontal edges

-Taillight presence; +4 if true

Vehicle verification according to the verification weight sum is applied differently according to the vehicle width. In the present invention, in the following case, a rule is established that a vehicle exists.

If the vehicle width is 60 pixels or more, then the total verification weight is 8 or more

-If the width of the vehicle is more than 40 pixels, then the total verification weight is 6 or more

-If the width of the vehicle is 25 pixels or more, then the total verification weight is 5 or more.

-If the vehicle's width is less than 25 pixels, the total weight is 4 or more.

18 is an image showing the front vehicle detected by the method proposed in the embodiment of the present invention.

What has been described above is only an exemplary embodiment of a front vehicle detection method using scene information of an image according to the present invention, and the present invention is not limited to the above-described embodiment, as claimed in the following claims, Without departing from the gist of the present invention, anyone of ordinary skill in the art will have the technical spirit of the present invention to the extent that various modifications can be made.

Claims (13)

In the front vehicle detection method using the scene information of the image,
Inputting an image acquired by a camera mounted on the driving vehicle;
Performing image preprocessing to remove noise of an input image and to improve edge information;
Determining a region of interest for vehicle candidate region extraction;
Extracting road and shadow areas by extracting road and shadow information;
Detecting horizontal and vertical lines of the vehicle and extracting a vehicle candidate region including the detected horizontal and vertical lines; And
Verifying whether a vehicle exists in the extracted vehicle candidate region; and detecting a front vehicle using scene information of an image.
The method of claim 1,
The performing of the image preprocessing,
Performing gray level image conversion when the input image is color;
Filtering for noise improvement and edge information improvement; And
And detecting horizontal edges and vertical edges using a Sobel edge detector.
The method of claim 2,
Wherein the filtering step is a front vehicle detection method using the scene information of the image, characterized in that the filtering using a rank order filter (Rank Order Filter).
The method of claim 1,
The determining of the region of interest,
Estimating a vanishing point;
Obtaining a horizontal line using the estimated vanishing point; And
Determining a region of interest for vehicle candidate region extraction, wherein the region of interest is determined as a space between a portion below the line 30 pixels higher than the horizontal line and at least 1/8 of the height of the entire image data. Vehicle detection method using the scene information.
5. The method of claim 4,
Estimating the vanishing point,
Performing image preprocessing to estimate vanishing points;
A huff line identification step of identifying a huff line by applying a hough transform on edges obtained in the image preprocessing, and then integrating similar huff lines having similar directions and adjacent positions; And
Estimating a vanishing point from the candidate huff line.
The method of claim 1,
Extracting the road area and the shadow area,
Determining a road statistical information ROI for extracting a road area and a shadow area;
Extracting road statistics information from the determined road statistical information ROI, comprising: obtaining an average m and a standard deviation v of gray level pixel values of the image image of the ROI; And
And extracting a road area and a shadow area based on the road statistics information.
The method according to claim 6,
Extracting the road area and the shadow area,
If any pixel gray level value (L) is in the range of [Equation 1] below,
[Formula 1]
Mean (m) -3 * standard deviation (v) <L <mean (m) + 3 * standard deviation (v)
The pixel is regarded as a road candidate pixel, and if the number of road candidate pixels in any region exceeds 60%, the corresponding area is extracted as a road region.
The method according to claim 6,
Extracting the road area and the shadow area,
If any pixel gray level value L satisfies [Equation 2] below,
[Formula 2]
L <mean (m)-4 * standard deviation (v)
A method of detecting a front vehicle using scene information of an image, which is regarded as a shadow candidate pixel and extracts a corresponding region as a shadow region when the shadow candidate pixel exceeds 50% in an arbitrary region.
The method of claim 1,
The extracting of the vehicle candidate region may include detecting horizontal and vertical lines from image data in which image preprocessing is performed; And
Extracting a vehicle candidate region by using the detected horizontal and vertical lines, wherein when a horizontal line exists and a vertical line is detected near the lower portion or the upper portion of the horizontal line, a quadrangle including the horizontal line and the vertical line And extracting the vehicle candidate region into a vehicle candidate region.
10. The method of claim 9,
The detecting of the horizontal and vertical lines may be performed by detecting a edge image obtained by applying a Sobel edge mask by applying a Hough transform on the binarized edge binarized image.
The method of claim 1,
The vehicle candidate region verification step may include:
Checking whether a shadow area exists within the vehicle candidate area;
Checking whether a road area exists at a lower portion of the vehicle candidate area;
Identifying additional verification conditions; And
And verifying whether the vehicle exists in the vehicle candidate region according to the verification weight sum by applying a weight to each verification result.
The method of claim 11,
The additional verification condition is any one or a combination of two or more of the presence or absence of wheels in the vehicle candidate region, the presence of symmetry, the horizontal edge distribution and the presence of a tail light, the front vehicle detection method using the scene information of the image.
The method of claim 11,
The step of verifying whether the vehicle exists in the vehicle candidate region according to the verification weighted sum includes verifying whether the vehicle exists by setting the verification weighted sum total higher as the vehicle width increases, and verifying that the vehicle exists. Detection method.

Images