JP2012080497A - Parking frame detection device and parking frame detection method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately detect straight lines to be candidates of both left and right ends of a line of a parking frame, prevent misdetection of an unrelated straight line as both ends of a line of the parking frame, and accurately detect the parking frame.SOLUTION: A parking frame detection device of the present invention comprises: imaging means that images the rear of a vehicle; edge extraction means that extracts edges from the imaged rear image; image conversion means that converts the image with the edges extracted therefrom into an overhead view image; area division means that divides the converted overhead view image into left and right areas; straight line detection means that detects straight lines in each area by Hough transform from the divided left and right areas; first determination means that determines whether the detected straight line is an end of a line having a width; and second determination means that creates a combination of two straight lines for each of the left and right areas from plural straight lines determined to be an end of a line having a width, and determines whether the combination is both ends of a line having a width.

Description

  The present invention relates to a parking frame detection device and a parking frame detection method, and more particularly to a parking frame detection device and a parking frame detection method for detecting a parking frame in a parking lot by an image processing technique from an image behind a vehicle imaged by an imaging means. It is.

When the vehicle is retracted and put into the parking frame, the image behind the vehicle imaged by the imaging means is converted into an image corrected for distortion and an overhead view image looking down from directly above by the image processing technology, to the driver. The parking assistance device to be presented has been put into practical use. The distortion-corrected image by the parking assist device is an image that is effective for the driver when the vehicle approaches the parking frame, because the line of the parking frame appears as a straight line while securing the surrounding visual field. On the other hand, the bird's-eye view image by the parking assistance device is effective when approaching the parking position because it is easy to grasp the sense of distance in the area close to the vehicle and the relationship between the line of the parking frame and the vehicle is intuitively easy to understand. As a method of detecting a parking frame from a captured image behind the vehicle, it is possible to extract a feature point of a straight line by binarization or edge extraction, conversion to a bird's-eye view image looking down from above, Hough transform (Hough transform) Processing such as straight line extraction by, determination of the width between lines of parking frames is used. (Patent Document 1 and Patent Document 2)

JP 2008-285083 A JP 2008-2137744 A

By the way, in the image processing technology for detecting the line of the parking frame from the image behind the vehicle imaged by the imaging means, it is good if only the white line of the parking frame is shown in the image, but there are various in an actual parking lot. There is a unique situation.
For example,
・ Line color of parking frame is other than white (yellow, orange, etc.)
-Lines other than parking frame lines (shadows, dirt on the ground, bodies of other vehicles, etc.)-Parking frame lines are not clearly visible (thin, discontinuous, or disappearing)
There are situations like this.
Under such circumstances, the above-described conventional method cannot be used, so that the parking frame cannot be detected with high accuracy.

  The present invention can accurately detect candidate straight lines at the left and right ends of the line of the parking frame, can prevent erroneous detection of an irrelevant straight line as both ends of the line of the parking frame, and can accurately detect the parking frame. An object of the present invention is to provide a parking frame detection device and a parking frame detection method that can be used.

  The present invention relates to an image pickup means for picking up the rear of a vehicle, an edge extraction means for extracting an edge from a rear image picked up by the image pickup means, and an image conversion for converting an image from which the edge is extracted by the edge extraction means into an overhead image. Means, an area dividing means for dividing the overhead image converted by the image converting means into left and right areas, and a straight line detecting means for detecting a straight line for each area from the left and right areas divided by the area dividing means by Hough transform And a first determination unit that determines whether or not the straight line detected by the straight line detection unit is an end of a line having a width, and the first determination unit determines that the line is an end of a line having a width. A second determination unit that forms a combination of two straight lines for each of the left and right regions from the plurality of straight lines and determines whether or not the combination is at both ends of a line having a width. And features.

Since the parking frame detection device of the present invention divides the bird's-eye view image into left and right regions, it can accurately detect straight lines that are candidates for the left and right ends of the line of the parking frame.
In addition, since the parking frame detection device of the present invention determines whether or not the straight line is an end of a line having a width, it can prevent erroneous detection of an irrelevant straight line as both ends of the line of the parking frame.
Furthermore, the parking frame detection device of the present invention creates a combination of two straight lines from a plurality of straight lines, and determines whether this combination is at both ends of a line having a width. It is possible to prevent erroneous detection of both ends of the line.
Thereby, the parking frame detection apparatus of this invention can detect a parking frame accurately.

FIG. 1 is a system configuration diagram of a parking frame detection device. (Example) FIG. 2 is a flowchart of parking frame detection of the parking frame detection device. (Example) FIG. 3 is a diagram showing an image captured by the camera of the imaging means. (Example) FIG. 4 is a diagram illustrating the overhead view image conversion of the edge. (Example) FIG. 5 is a diagram showing straight line detection by Hough transform. (Example) FIG. 6 is a diagram showing extraction of straight line candidates. (Example) FIG. 7 is a view showing an overhead image converted from a luminance image. (Example) FIG. 8 is a diagram illustrating verification of luminance change of a straight line candidate. (Example) FIG. 9 is a diagram showing line candidate extraction. (Example) FIG. 10 is a diagram illustrating the verification of the luminance change of each straight line candidate in the left region. (Example) FIG. 11 is a diagram showing verification of luminance change of each straight line candidate in the right region. (Example) FIG. 12 is a flowchart of straight line candidate combination verification. (Example) FIG. 13 is a diagram showing verification of each straight line candidate combination in the left region. (Example) FIG. 14 is a flowchart of parking frame determination. (Example) FIG. 15 is a diagram illustrating determination of a parking frame. (Example) FIG. 16 is a system configuration diagram of the parking assistance device. (Example) FIG. 17 is a diagram showing an image captured by the camera of the imaging means. (Example) FIG. 18 is a diagram illustrating automatic switching from a distortion-corrected image to an overhead image. (Example) FIG. 19 is an operation flowchart of the parking assist device. (Example)

  Embodiments of the present invention will be described below with reference to the drawings.

1 to 19 show an embodiment of the present invention.
The parking assistance device that displays an image and presents it to the driver when the vehicle is moved backward and put into the parking frame switches between the distortion correction image and the overhead image and displays them. The driver can switch between the distortion-corrected image and the bird's-eye view image by himself / herself, but when driving backwards, driving tasks such as checking the surroundings and operating the steering wheel are concentrated. There is annoyance.
The applicant of the present invention has already filed an application for a parking assistance apparatus that employs a method of automatically switching between the distortion-corrected image and the overhead view image in order to eliminate the troublesome operation of switching the image. As shown in FIG. 17, the parking assist device detects the left and right lines 3 _L and 3 _R of the parking frame 2 of the parking lot 1 from the image G1 behind the vehicle imaged by the imaging means, and the vehicle and the line 3 _L If the 3 and _R are in parallel, the line ₃ L · ₃ detects the rear end ₃ L a · _{3 R} a of _R, the vehicle and the line trailing ₃ L a · ₃ distance between _R a is a threshold When the value falls below the value, the image is automatically switched. Reference numerals 4 _L and 4 _R are left and right wheel stops of the parking lot 1. The color of the line 3 _L · 3 _R parking space 2 is usually a white, there is a yellow, orange or the like in addition to white.

As shown in FIG. 16, the parking assistance device 5 that automatically switches images includes a camera 6 that captures an image of the rear of the vehicle, a monitor 7 that displays an image, and a vehicle captured by the camera 6. Support control means 8 for displaying a rear image on the monitor 7 is provided. The camera 6 includes an imaging device such as an ultra-wide angle back camera attached to the rear part of the vehicle. The monitor 7 is composed of a display device such as a monitor of a car navigation system arranged in the vehicle interior of the vehicle.
The support control means 8 captures and records an image (see FIG. 17) including the parking frame 2 behind the vehicle imaged by the camera 6, and an image G1 recorded in the image recording means 9. The image switching determination means 10 that determines the switching of the display image from the information and outputs it to the subsequent distortion correction image conversion means 11 or the overhead image conversion means 12, and the distortion of the input image G1 by correcting the distortion of the input image G1 The distortion-corrected image converting means 11 for converting the image G1, the overhead image converting means 12 for converting the input image G1 into the overhead image G3, and the distortion-corrected image G2 or the overhead image G3 are switched and displayed (see FIG. 18). And an image output means 13 for outputting to the monitor 7.

As shown in FIG. 19, when the operation program is started (101), the parking assist device 5 captures and records the image G1 behind the vehicle imaged by the camera 6 in the image recording means 9 (102). The position and angle of the line 3 _L · 3 _R of the parking frame 2 are calculated from the recorded image G 1 (103), and it is determined whether the vehicle and the line 3 _L · 3 _R are parallel (104).
If this determination (104) is NO, the distortion corrected image conversion means 11 converts the image into a distortion corrected image G2 (105), outputs the distortion corrected image G2 to the monitor 7 (106), and ends the program ( 107). If the determination (104) is YES, the image switching determination means 10 detects the rear end 3 _L a · 3 _R a of the line, detects the loop stop 4 _L · 4 _R , and counts the vehicle speed pulse (108), It is determined whether to switch the image output to the monitor 7 by the image output means 13 (109).
If this determination (109) is NO, the distortion corrected image converting means 11 converts the image G1 into a distortion corrected image G2 (105), and the image output means 13 outputs the distortion corrected image G2 to the monitor 7 (106). The program is terminated (107). On the other hand, when this determination (109) is YES, the overhead image conversion unit 12 converts the image G1 into the overhead image G3 (110), and the image output unit 13 outputs the image output to the monitor 7 from the distortion correction image G2. The image G3 is switched and output (106), and the program is terminated (107).

In the parking assistance device 5, in order to use it as information for assisting parking, as shown in FIGS. 1 to 15, the parking frame 2 is detected from the captured image G behind the vehicle, and the parking frame detection device and the parking frame are used. It is detected by the detection method. The parking assist device 5 uses the information of the parking frame 2 detected by the parking frame detection device and the parking frame detection method, for example, image switching determination by the image switching determination unit 10, distortion correction image conversion unit 11, and overhead image conversion unit. 12 for image conversion.
As shown in FIG. 1, a parking frame detection device 14 that detects a parking frame from an image behind the vehicle includes a camera 15 as an imaging unit that captures the rear of the vehicle, and the parking frame 2 from the rear image captured by the camera 15. The detection control means 16 which detects this is provided.
The detection control means 16 captures and records the image G4 behind the vehicle imaged by the camera 15 (see FIG. 3), and the edge e from the rear image G4 recorded in the image recording means 17. Edge extracting means 18 for extracting, image converting means 19 for converting an image from which the edge e is extracted by the edge extracting means 18 into an overhead image G5 (see FIG. 4), and an overhead image converted by the image converting means 19 a region dividing means 21 for G5 divided into left and right areas ₂₀ L · 20 _R (see FIG. 5), region ₂₀ by Hough transformation from divided left and right regions ₂₀ L · 20 _R by the region dividing means 21 L · 20 _R Yes a straight line detecting means 23 for detecting a straight line ₂₂ L · _{22 R} (see FIG. 5), respectively, the straight lines ₂₂ L · 22 _R width detected by the straight line detecting unit 23 for each Left and right from a plurality of straight lines 22 _L and 22 _R determined by the first determination means 24 to be ends of a line having a width. And a second determination means 25 for making a combination of two straight lines 22 _L and 22 _R for each of the regions 20 _L and 20 _R and determining whether or not the combination is at both ends of a line having a width. ing.
Since the parking frame detection device 14 of this embodiment is built in the parking assistance device 5, the camera 15 and image recording means 17 of the parking frame detection device 14 are replaced with the camera 6 and image recording means of the parking assistance device 5. 9 can be substituted.
Parking space detecting unit 14 from the image G4 to extract an edge e, converts the image obtained by extracting the edge e in the overhead image G5, it divides the bird's-eye image G5 in left and right regions 20 L _· 20 _R, left and right region ₂₀ L · 20 by Hough transformation from _R to detect an area ₂₀ L · 20 respectively each _R linear ₂₂ L · 22 _R, parking in line (this example the detected straight line ₂₂ L · 22 _R has a width It is determined whether or not it is an end of the line 3 _L · 3 _R ) of the frame 2, and the left and right regions 20 _L · 20 _R from the plurality of straight lines 22 _L · 22 _R determined to be the ends of the line having a width. A combination of two straight lines is created every time, and it is determined whether or not the combination is at both ends of a line having a width.

Next, detection of a parking frame by the parking frame detection device 14 will be described. As shown in FIG. 2, when the parking frame detection program is started (201), the parking frame detection device 14 captures and records the image G4 behind the vehicle imaged by the camera 15 in the image recording means 17 (FIG. 2). 3) (202), the vertical edge e is extracted from the recorded image G4 and converted into an overhead image G5 (see FIG. 4) (203). In the parking frame detection device 14, the image is switched from the image G 4 to the line 3 _L · 3 in order to switch the image while the vehicle is substantially parallel to the left and right lines 3 _L · 3 _R of the parking frame 2 at the final stage of the parking operation. Only the vertical edge e extending in the _R direction may be extracted.
The candidates of the left and right straight lines 22 _L and 22 _R are detected by the Hough transform process (see FIG. 5) using the edge e point on the overhead image G5 by the transformation (203) as a vote point (204). At this time, the bird's-eye view image G5 is divided into right and left half regions 20 _L and 20 _R , and candidates for straight lines 22 _L and 22 _R are detected in the left and right regions 20 _L and 20 _R by Hough transform.
As shown in FIG. 5, the detection (204) of the straight lines 22 _L and 22 _R is performed from the coordinate values x and y of each candidate point p (x, y) of the overhead image G5.
ρ = x ^* sin (θ) + y ^* cos (θ)
The value of ρ is calculated from θmin to θmax and is voted on the ρ-θ space.
The locations P _L1 (ρ _L1 , θ _L1 ) and P _R1 (ρ _R1 , θ _R1 ) having the highest number of votes in each of the left and right regions 20 _L and 20 _R of the bird's-eye view image G5 are defined as the first candidate straight line (hereinafter referred to as “ _R” This is referred to as a “first straight line.” 22 _L1 · 22 _R1 and the second highest point is P _L2 (ρ _L2 , θ _L2 ) and P _R2 (ρ _R2 , θ _R2 ) as a second candidate straight line (hereinafter referred to as “first straight line”). This is referred to as a “second straight line.” 22 _L2 · 22 _R2 , and the third highest point is P _L3 (ρ _L3 , θ _L3 ) and P _R3 (ρ _R3 , θ _R3 ) as a third candidate straight line (hereinafter referred to as “second straight line”). This is described as “third straight line.”) 22 _L3 · 22 _R3 (see FIG. 6).
Following detection of the candidates for the straight lines 22 _L and 22 _{R in} the left and right regions 20 _L and 20 _R (204), the luminance image is converted into an overhead image G5 (see FIG. 7) (205). The image is converted by calculating which pixel on the overhead image G5 each pixel on the image G4 captured by the camera 15 corresponds to. At this time, the calculation result may be stored in a table in advance, and the table may be referred to at the time of conversion.

Each straight line candidate (the straight line candidate is not a high probability of being a straight line, but is a straight line, meaning a candidate for edges e at both ends in the width direction of the lines 3 _L and 3 _R from the overhead image G5 of the converted luminance. ) Is verified (see FIG. 8) (206). In the verification (206) as the first determination, as shown in FIG. 8, the neighboring areas A _L on the left and right sides of the first to third straight lines 22 _{L1 to} 22 _L3 and 22 _{R1 to} 22 _R3 , which are straight line candidates, respectively. _Calculate the luminance average L / R of AR.
If the obtained absolute value of the difference between the luminance average L · R is less than the threshold (th) (| luminance average L−luminance average R | <th), it is not determined as a straight line candidate as NG, but the absolute value of the difference between the luminance average L · R Is equal to or greater than the threshold value (th) (| brightness average L 1 luminance average R | ≧ th), a straight line candidate is determined as OK. Further, from the magnitude relationship of the luminance average L · R, if the luminance average L is less than the luminance average R (luminance average L <luminance average R), the left edge is set, and the luminance average L is equal to or greater than the luminance average R (luminance average L ≧ luminance average). R) is determined as the right edge, and it is determined whether the line 3 _L · 3 _R having the width of the parking frame 2 is the left edge or the right edge.
The luminance change verification (206), which is the first determination, will be described in more detail with reference to FIGS. As shown in FIG. 9, in the left and right regions 20 _L and 20 _{R obtained} by dividing the overhead view image G5 into the right half and the left half, the first to third straight lines 22 _{L1 to} 22 _L3 and the first to third are respectively used as straight line candidates. A case where the straight lines 22 _{R1 to} 22 _R3 are extracted will be described.
As shown in FIG. 10C, the third line 22 _L3 of the left region 20 _L is excluded from the line candidates because the absolute value of the difference between the luminance average L · R is less than the threshold value (th). As shown in FIGS. 10A and 10B, the first straight line 22 _L1 and the second straight line 22 _L2 of the left region 20 _L have an absolute value of the difference between the luminance average L · R equal to or greater than a threshold value (th). Straight line candidates. At this time, it is assumed that the first straight line 22 _L1 is the left edge (black-white) and the second straight line 22 _L2 is the right edge (white-black) because of the magnitude relationship between the absolute values of the differences between the luminance averages L and R.
As shown in FIGS. 11A to 11C, the first straight line 22 _R1 to the third straight line 22 _R3 of the right region 20 _R all have an absolute value of the difference between the luminance average L · R equal to or greater than a threshold value (th). Therefore, it will be a straight line candidate. At this time, the first straight line candidate 22 _R1 is the right edge (white-black), the second straight line 22 _R2 is the left edge (black-white), and the third straight line from the magnitude relationship of the absolute value of the difference between the luminance average L · R. 22 _R3 is the left edge (black-white).

As shown in FIG. 2, following the verification (206) of the luminance change in each line candidate, first to third combination of straight lines ₂₂ L1 _{through 22 L3} of the left region _{20 L (22 L1 -22 L2,} 22 L1 -22 _{L3, 22} L2 _{-22 L3} validity of), the right area 20 first to third straight lines _{R 22} R1 _{through 22 R3} combinations _{(22 R1 -22 R2, 22 R1} -22 R3, 22 R2 -22 R3) The validity is verified and it is determined whether each combination is OK or NG (207). Further, a straight line candidate combination is determined from the verification result of each combination (208).
In the combination verification (207) as the second determination, as shown in FIG. 12, when the verification program is started (301), the distance ρ from the origin of the two straight line candidates is smaller. The left edge has the larger distance ρ from the origin of the two straight line candidates as the right edge (302), and it is determined whether the left-right relation of ρ and the left-right relation of the luminance average L · R match (303). If this determination (303) is YES, it is determined whether the absolute value of the difference in the distance ρ is within a certain range (304). If this determination is YES, it is determined whether the absolute value of the difference between the angles θ of the two straight line candidates is equal to or smaller than a certain value (305). If this determination (305) is YES, the combination is OK (306) and the program is terminated (307).
On the other hand, if the determinations (303), (304), and (305) are NO, the combination is determined as NG (308), and the program is terminated (307).
In the verification of this combination (207), for example, in the case of the left region 20 _L, the first straight line _{22 L1} - if the determination of the combination of the second linear _{22 L2} OK, first straight line _{22 L1} - second linear _{22 L2} Is a straight line candidate. If the determination of the first straight line 22 _{L1 -the} second straight line 22 _L2 is NG, the combination of the first straight line 22 _{L1 -the} third straight line 22 _L3 is determined. If the determination of the first straight line 22 _{L1 to} the third straight line 22 _L3 is OK, the first straight line 22 _{L1 to} the third straight line 22 _L3 are set as straight line candidates. If the determination of the first straight line 22 _{L1 -the} third straight line 22 _L3 is NG, the combination of the second straight line 22 _{L2 -the} third straight line 22 _L3 is determined. If the determination of the second straight line 22 _{L2 to} the third straight line 22 _L3 is OK, the second straight line 22 _{L2 to} the third straight line 22 _L3 are set as straight line candidates. If the determination of the second straight line 22 _{L2 -the} third straight line 22 _L3 is NG, it is determined that there is no straight line candidate. Also in the case of the right region 20 _R, similarly determines first through combination of the third straight line ₂₂ R1 _{through 22 R3,} determining the presence or absence of line candidates.

In the left region 20 _L of FIG. 9, the first straight line _{22 L1,} since the second linear _{22 L2} remained, first straight line _{22 L1} - verifies the combination of the second straight line _{22 L2,} right area 20 _R of FIG. 11 Since the first straight line 22 _R1 to the third straight line 22 _R3 remain, the first straight line 22 _{R1 -the} second straight line 22 _R2 , the first straight line 22 _{R1 -the} third straight line 22 _R3 , the second straight line 22 _{R2 -the} third The validity of each combination of the straight line 22 _R3 is verified.

• Does the right / left relation of ρ match the right / left relation of the luminance average L / R?
ρ represents the distance from the origin to the straight line (see FIG. 5). Since the origin is at the left end, the smaller value of ρ (shorter distance) is on the left, and the larger value of ρ (longer distance) is on the right. As shown in FIG. 13, in the left region 20 _L , the value of ρ2 of the second straight line 22 _L2 on the right side far from the origin is larger than the value of ρ1 of the first straight line 22 _L1 on the left side near the origin.
In the left region 20 _L shown in FIG. 10, the relationship between the first straight line 22 _L1 and the second straight line 22 _L2 is that the luminance difference direction (the first straight line 22 _L1 is left, the second straight line 22 _L2 is right) and the left and right of ρ The relationship (the first straight line 22 _L1 is on the left and the second straight line 22 _L2 is on the right) matches.
In the right region 20 _R shown in FIG. 11, the relationship between the first straight line 22 _R1 and the second straight line 22 _R2 is that the luminance difference direction (the first straight line 22 _R1 is right and the second straight line 22 _R2 is left) and the left and right of ρ The relationship (first straight line 22 _R1 is left, second straight line 22 _R2 is right) does not match. The first straight line _{22 R1,} the relationship of the third straight line _{22 R3} is the direction of the luminance difference (first straight line _{22 R1} is right, the third straight line _{22 R3} is left) lateral relationship (first straight line _{22 R1} and ρ is the right, The third straight line 22 _R3 matches the left). Second linear _{22 R2,} the relationship of the third straight line _{22 R3} is the direction of the luminance difference (second straight line _{22 R2} is left, the third straight line _{22 R3} is left) and NG is the same.

・ Is the absolute value of ρ difference within a certain range?
Since the absolute value of the difference of ρ is the distance between straight lines (see FIG. 5), it represents the line width (line thickness) of the lines 3 _L and 3 _R of the parking frame 2. Check if the distance is within a certain range. Decide the minimum and maximum values for the line widths 3 _L and 3 _R of the parking frame 2 to be detected in advance.
Minimum value <absolute value of difference of ρ <maximum value → OK
And

・ Is the absolute value of the θ difference smaller than a certain value?
The absolute value of the difference in θ represents the angle difference between straight lines (see FIG. 5). Check if the angle difference is below a certain value (parallel). Since the line width (line thickness) of the line 3 _L · 3 _R of the parking basket 2 is constant, it is not the line 3 _L · 3 _{R when} it is not parallel.

As shown in FIG. 2, after the straight line candidate combination is determined (208), the presence / absence of the parking frame 2 is determined (209). In the determination of the parking frame 2, as shown in FIG. 14, when the determination program is started (401), the left and right areas 20 _L and 20 _R are straight line candidates indicating the lines 3 _L and 3 _R of the parking basket 2. In FIG. 15, it is determined whether or not there exists the first straight line 22 _{L1 -the} second straight line 22 _L2 in the left region 20 _L and the first straight line 22 _{R1 -the} third straight line 22 _{R3 in the} right region 20 _R (402).
If the determination (401) is YES, the line _{3 L} and the line ₃ the difference between the distance ρ from the origin of the _R (in FIG. 15, the second straight line _{22 L2,} the distance between the third straight line _{22 R3)} is constant It is determined whether it is within the range (403). When the determination (403) is YES, the difference in the angle θ to the origin of the line _{3 L} and the line _{3 R} (see FIG. 5) is equal to or less than a predetermined range value (404). If this determination (404) is YES, it is assumed that there is a parking frame 2 (405), and the program is terminated (406).
On the other hand, if the determinations (402), (403), and (404) are NO, it is determined that the parking frame 2 does not exist (407), and the program ends (406).

Determination of the presence or absence of the parking frame 2 (209) in, in each of the left and right regions ₂₀ L · 20 _R, a combination of the straight line _{22 L (22 L1 -22 L2,} 22 L1 -22 L3, 22 L2 -22 L3) and linear 22 _R combination _{(22 R1 -22 R2, 22 R1} -22 R3, 22 R2 -22 R3) of, when a combination of OK is present, in which it is determined whether the parking space 2.
Of the combinations of the first to third straight lines 22 _{L1 to} 22 _L3 of the left region 20 _L , the right line (second straight line 22 _L2 in FIG. 15) is the first to third straight lines 22 _R1 of the right region 20 _R. left line of the combination through 22 _R3 (in Fig. 15 the third linear _{22 R3)} selected.
It is confirmed whether the difference in the distance ρ from the origin between the second straight line 22 _L2 of the left region 20 _L and the third straight line 22 _R3 of the right region 20 _R is within a certain range. Decide the minimum and maximum values for the width of lines 3 _L and 3 _R of the parking frame 2 to be detected in advance.
Minimum value <difference of ρ <maximum value → OK
And
Furthermore, to confirm that it is the difference between the angle θ to the origin of the third straight line _{22 R3} of the second straight line _{22 L2} and the right region 20 _R of the left region 20 _L is smaller than a predetermined value (parallel). If the difference of the angle θ is equal to or less than a certain value, the left and right lines 3 _L and 3 _R of the parking frame 2 are assumed to be parallel.

  As shown in FIG. 2, after determining whether there is a parking frame 2 (209), the program ends (210). Information on the determination result of the parking frame 2 by the parking frame detection device 14 is used in the parking support device 5 for image switching determination and the like.

Thus, the parking frame detection device 14 includes the camera 15, the image recording means 17, the edge extraction means 18, the image conversion means 19, the area dividing means 21, the straight line detection means 23, and the first determination means. 24 and the second determination means 25, the edge e is extracted from the image G4, the image from which the edge e is extracted is converted into an overhead image G5, and the overhead image G5 is converted into the left and right regions 20 _L and 20. _The straight lines 22 _L and 22 _R are detected for each of the areas 20 _L and 20 _R by the Hough transform from the left and right areas 20 _L and 20 _R , and the detected straight lines 22 _L and 22 _R have a width. It is determined whether or not it is an end, and a combination of two straight lines is created for each of the left and right regions 20 _L and 20 _R from a plurality of straight lines 22 _L and 22 _R determined to be ends of a line having a width, Whether this combination is at both ends of a line with width Determines whether or not.
Since this parking frame detection device 14 divides the bird's-eye view image G5 into the left and right regions 20 _L and 20 _R , the parking frame detection device 14 can accurately detect candidate straight lines at the left and right ends of the lines 3 _L and 3 _R of the parking frame 2. it can. Further, the parking space detecting unit 14, to determine whether the straight line 22 _R of the straight line 22 _L and the left region 20 _R of the right region 20 _L is an end of a line having a width, independent linear 22 _L · It is possible to prevent 22 _R from being erroneously detected as both ends of the line 3 _L · 3 _R of the parking frame 2. Further, the parking frame detection device 14 creates a combination of two straight lines from the plurality of straight lines 22 _{L1 to} 22 _L3 and creates a combination of two straight lines from the plurality of straight lines 22 _{R1 to} 22 _R3. In order to determine whether or not it is at both ends of the line 3 _L · 3 _R having the width of the parking frame 2, it is possible to prevent erroneous detection of an irrelevant straight line as both ends of the line 3 _L · 3 _R of the parking frame 2. it can.
Thereby, this parking frame detection apparatus 14 can detect the parking frame 2 with high accuracy.

Moreover, the parking frame detection apparatus 14 can have the following effects.
The first determination means 24 sets areas A _L and A _R having a constant width in the vicinity of both sides of each straight line 22 _L and each straight line 22 _R detected by the Hough transform, and each of the areas A _L and A _R The luminance average of the pixels is calculated, and the difference in luminance average between the areas A _L and A _R on both sides of each straight line 22 _L and each straight line 22 _R is calculated. When the absolute value of this difference is less than a predetermined value, The straight line _22L and the straight line _22R are not subject to subsequent determination.
Thereby, the parking frame detection apparatus 14 removes a straight line that does not apply to either end of the line 3 _L or 3 _R having the width of the parking frame 2, thereby removing an unrelated straight line from the single line 3 of the parking frame 2. it can be prevented from being erroneously detected as the left and right ends of the _L or line 3 _R.

The second determination unit 25, the average luminance magnitude relationship between the sides of the area _A L · _{A R} in each straight line 22 _L and the straight lines 22 _R, which side of the line the straight line 22 _L and the straight line 22 _R has a width If the two straight lines 22 _L and the two straight lines 22 _{R which} are arbitrarily selected are the ends on the same side of the line having the width, the two straight lines 22 _L are detected. It is not used for subsequent determination of the combination of _L and two lines 22 _R.
As a result, the parking frame detection device 14 removes a combination of two straight lines that do not correspond to both ends of the line 3 _L or 3 _R having a width, thereby removing an unrelated straight line from the single line 3 _L or the parking frame 2. it can be prevented from being erroneously detected as the left and right ends of the lines 3 _R.

Also, the second determination unit 25, the positional relationship between the positional relationship and the two straight lines 22 _R of the two straight lines 22 _L detected from the magnitude relation of the luminance average, on both sides of a line straight line having a width even when appropriate to the end, when in the positional relationship improbable as opposite ends of a line straight line having a width, a combination of subsequent determination of combination and two straight 22 _R of the straight line 22 _L of these two Not targeted.
Thus, the parking space detecting unit 14, by removing the two straight lines 22 _L combinations combinations and two straight 22 _R in positional relationship improbable, unrelated straight parking frame 2 one line 3 it can be prevented from being erroneously detected as the left and right ends of the _L or line 3 _R.

Further, the second determination unit 25, when not in the range where the distance between the distance and the two straight lines 22 _R between two straight lines 22 _L is preset, linear 22 of the two _L the combination of the combination and the two straight lines 22 _R not subject to subsequent determined.
As a result, the parking frame detection device 14 removes a combination of two straight lines that cannot be used as the width of the lines 3 _L and 3 _R of the parking frame 2, thereby converting an irrelevant straight line into the single line 3 _L or the parking frame 2. it can be prevented from being erroneously detected as the left and right ends of the lines 3 _R.

Furthermore, the second determination unit 25, when not within the angular difference between the angular difference and the two straight lines 22 _R between two straight lines 22 _L is set in advance, the two It is not used for subsequent determination of the combination and the two combination of straight lines 22 _R of the straight line 22 _L.
As a result, the parking frame detection device 14 _removes a combination of the two straight lines 22 _L or the combination of the straight lines 22 _R whose line widths are changed, thereby removing an unrelated straight line from the single line 3 _L of the parking frame 2. or erroneously detected as the left and right ends of lines 3 _R can be prevented.

Further, the edge extracting means 18 extracts the vertical edge e from the image G4 of the camera 15.
Thereby, when detecting the parking frame 2 when the vehicle is parallel to the left and right lines 3 _L and 3 _R of the parking frame 2, the parking frame detection device 14 can reduce the amount of calculation of edge extraction.

  The present invention can accurately detect candidate straight lines at the left and right ends of the line of the parking frame, can prevent erroneous detection of an irrelevant straight line as both ends of the line of the parking frame, and can accurately detect the parking frame. Therefore, the present invention can be used for detecting a linear figure without being limited to a parking frame.

DESCRIPTION OF SYMBOLS 1 Parking lot 2 Parking frame 3 _L left line 3 _R right line 5 Parking assistance apparatus 14 Parking frame detection apparatus 15 Camera 16 Detection control means 17 Image recording means 18 Edge extraction means 19 Image conversion means 20 _L Left area 20 _R Right area 21 Area dividing means 22 _L Left straight line 22 _R Right straight line 23 Straight line detecting means 24 First determining means 25 Second determining means

Claims (8)

Imaging means for imaging the rear of the vehicle;
Edge extraction means for extracting an edge from a rear image captured by the imaging means;
Image conversion means for converting the image from which the edge has been extracted by the edge extraction means into an overhead image;
Area dividing means for dividing the overhead image converted by the image converting means into left and right areas;
Straight line detecting means for detecting straight lines for each area by Hough transform from the left and right areas divided by the area dividing means,
First determination means for determining whether or not the straight line detected by the straight line detection means is an end of a line having a width;
A combination of two straight lines is created for each of the left and right regions from a plurality of straight lines determined to be the ends of a line having a width by the first determining means, and whether or not this combination is both ends of a line having a width. A parking frame detection device comprising: a second determination unit that determines whether or not.   The first determination means sets an area with a certain width near both sides of each straight line detected by the Hough transform, calculates the average luminance of the pixels in each area, and calculates the average luminance of the areas on both sides in each straight line. The parking frame detection device according to claim 1, wherein a difference is calculated, and when the absolute value of the difference is less than a predetermined value, the straight line is not subjected to subsequent determination.   The second determining means detects whether the straight line corresponds to an end of a line having a width from the luminance average magnitude relationship between the areas on both sides of each straight line, and two arbitrarily selected straight lines 3. The parking frame detection device according to claim 2, wherein when the two are ends on the same side of a line having a width, the combination of the two straight lines is not a target for subsequent determination.   The second determination means is configured so that, even when the positional relationship between the two straight lines detected from the luminance average magnitude relationship corresponds to the ends on both sides of the line having the width, the straight line has both sides of the line having the width. The parking frame detection device according to claim 3, wherein the combination of the two straight lines is not a target for subsequent determination when the position is in an impossible positional relationship.   The second determination unit is characterized in that, when the distance between two straight lines is not within a preset range, the combination of the two straight lines is not subject to subsequent determination. The parking frame detection apparatus as described in any one of Claims 1-4.   The second determination unit is characterized in that, when an angle difference between two straight lines is not within a preset range, the combination of the two straight lines is not subjected to subsequent determination. The parking frame detection device according to any one of claims 1 to 5.   The parking frame detection apparatus according to claim 1, wherein the edge extraction unit extracts a vertical edge from an image. Extract an edge from the image, convert the extracted image into an overhead image,
This overhead image is divided into left and right areas,
Detect straight lines for each area by Hough transform from the left and right areas,
Determine whether the detected straight line is the end of a line having a width;
A combination of two straight lines is created for each of the left and right regions from a plurality of straight lines determined to be the ends of a line having a width, and it is determined whether or not this combination is at both ends of a line having a width. Parking frame detection method.

Images