WO2016147899A1

WO2016147899A1 - Device for hanger detection by image processing

Info

Publication number: WO2016147899A1
Application number: PCT/JP2016/056715
Authority: WO
Inventors: 庭川　誠; 勇介渡部; 匠朗川畑
Original assignee: 株式会社明電舎
Priority date: 2015-03-13
Filing date: 2016-03-04
Publication date: 2016-09-22
Also published as: TW201636572A; MY186370A; SG11201707279TA; TWI597471B; JP2016168937A; CN107428261B; CN107428261A; JP6427797B2

Abstract

This device for hanger detection by image processing detects hangers (6) in an image, and is equipped with: a line sensor camera (2) that captures an image of hangers (6) from the roof of a vehicle (1), the hangers (6) being installed with equal spacing and support a trolley line (4); and an image processing unit (3) that is installed within the vehicle (1) and analyzes the image obtained by the line sensor camera (2). The device for hanger detection is configured such that the image processing unit (3) is provided with: a gray scale-top hat processing unit (31a) which sequentially subjects each line sensor image input from the line sensor camera (2) to the gray scale-top hat processing in the imaging line direction and time direction of the line sensor camera (2) and acquires a gray scale-top hat image where the hangers (6) in the image are highlighted; and a hanger detection unit (31b) which detects the hangers (6) from the gray scale-top hat image.

Description

Hanger detection device by image processing

The present invention relates to a hanger detection device by image processing that detects a hanger by image processing and detects a travel position of a train car and an angle abnormality of the hanger.

First, the necessity of detecting hangers will be described.
A railway operator records a train track while traveling and specifies the traveling position of the train from the recorded image. Conventionally, there are GPS and ATC technologies for identifying the location of trains. However, these technologies are accurate enough not to interfere with train operation, but when train locations are identified from images. Therefore, it is required to grasp the traveling position more accurately.

In addition, normal hangers are installed vertically, but it is important to know the angle of the hangers because the current collection characteristics of the train deteriorate if they are inclined with respect to the vertical direction due to temperature changes. . Conventionally,

Patent Documents

1 and 2 are known as techniques for detecting hangers. Patent Document 1 is a patent that detects hangers from captured images by template matching processing or vertical edge detection processing. Patent Document 2 is a patent for detecting a hanger by detecting an intersection of two straight lines from a captured image by detecting a Hough transform process, a luminance histogram process, or two straight lines.

Japanese Patent No. 5402096 Japanese Patent No. 5402272 JP 2013-015336 A

The inventions disclosed in

Patent Documents

1 and 2 are capable of detecting hangers as long as they have a simple line configuration. However, as shown in FIG. It is difficult to detect the hanger 6 at a location where 5 intersects.
Patent document 1 detects a hanger by vertical edge detection. In general edge detection, a portion having a large differential value between 2 pixels is detected as an edge. The local detection of 2 pixels has a problem with many false detections.
Patent document 2 is a patent which detects a hanger by Hough transformation. The Hough transform finds a hanger using a relative vote amount between the feature of the entire image and the hanger feature, so it is difficult to detect the hanger 6 that appears small relative to the entire image as shown in FIG.

There is Patent Document 3 as a related patent. This is an invention for measuring wear of a trolley wire, and is a patent for robustly detecting a worn portion of a trolley wire by removing the background by GSTH processing (see FIGS. 4 and 5). The GSTH process of Patent Document 3 is characterized in that the illumination light on the lower surface of the trolley wire is reflected in white. Since there is no part which reflects white in the hanger which is the object of this case, it is difficult to detect the hanger using Patent Document 3.

Therefore, an object of the present invention is to provide a hanger detection device using image processing that can detect a hanger with higher accuracy.

A hanger detection device by image processing according to the first invention for solving the above-mentioned problems is
A line sensor camera for photographing a hanger that is installed at equal intervals and supports a trolley line from the roof of the vehicle, and an image processing unit that is installed inside the vehicle and analyzes an image obtained by the line sensor camera, A hanger detection device by image processing for detecting a hanger in an image,
The image processing unit
A GSTH processing unit that sequentially performs GSTH processing on a line sensor image input from the line sensor camera in a shooting line direction and a time direction of the line sensor camera to obtain a GSTH image in which the hanger in the image is emphasized; ,
And a hanger detector for detecting a hanger from the GSTH image.

Moreover, the hanger detection apparatus by the image processing which concerns on 2nd invention in 1st invention,
The GSTH processing unit obtains a first GSTH image by performing an expansion process on the line sensor image after performing a contraction process on the imaging line direction of the line sensor camera as the GSTH process, and acquiring the first GSTH image. A difference image is obtained by taking a difference between the image and the first GSTH image, and after the shrinkage process is performed on the difference image in the time direction, the expansion process is performed to emphasize the hanger in the line sensor image Get a second GSTH image,
The hanger detection unit detects the hanger from the second GSTH image.

Moreover, the hanger detection apparatus by the image processing which concerns on 3rd invention in 1st invention,
As the GSTH processing, the GSTH processing unit obtains a gsth image by performing only a contraction process on the line sensor image in the photographing line direction of the line sensor camera, and acquires a gsth image in the time direction on the gsth image. The GSTH image in which the hanger in the line sensor image is emphasized by performing an expansion process after performing
The hanger detection unit detects the hanger from the GSTH image.

Moreover, the hanger detection apparatus by the image processing which concerns on 4th invention is the invention in any one of 1st to 3rd.
The GSTH processing unit inverts the luminance value for the line sensor image, one line at a time for the shooting line direction of the line sensor camera, and the luminance value of the main component in the shooting line direction when the luminance value of the main component is light color. When the value is a dark color, the GSTH process is performed after the luminance inversion process that does not invert the luminance value.

According to the present invention, it is possible to extract only hangers that exist in the image in a predetermined direction and as a straight line that is thinner than a predetermined thickness, so that only normal hangers whose installation angle is within the control value can be detected with high accuracy. can do.

It is explanatory drawing which shows the example of installation of the hanger detection apparatus by the image process which concerns on Example 1 of this invention. It is a block diagram which shows the apparatus structure of the hanger detection apparatus by the image processing which concerns on this invention. It is explanatory drawing which shows the example of the original image for demonstrating GSTH process. It is explanatory drawing which shows an example of the shrinkage | contraction process in a GSTH process. It is explanatory drawing which shows an example of the expansion process in a GSTH process. It is explanatory drawing which shows the conversion image obtained as a result of carrying out the GSTH process of the original image shown in FIG. It is a flowchart which shows the flow of the hanger detection in Example 1 of this invention. It is a schematic diagram which shows an example of a line sensor image. It is a schematic diagram which shows the image which reversed the luminance value of the line sensor image shown in FIG. FIG. 10 is a schematic diagram illustrating an example in which the image shown in FIG. 9 is subjected to GSTH processing in a region long in the x direction. It is a schematic diagram which shows the image which took the difference of the image shown in FIG. 9, and the image shown in FIG. FIG. 12 is a schematic diagram illustrating an example in which the image shown in FIG. 11 is subjected to GSTH processing in a region long in the y direction. It is a flowchart which shows the flow of the hanger detection in Example 2 of this invention.

Hereinafter, a hanger detection apparatus using image processing according to the present invention will be described with reference to the drawings, but the present invention is not limited to the following embodiments.
In the following description, the process denoted as GSTH process indicates that the expansion process illustrated in FIG. 5 is performed after the contraction process illustrated in FIG. 4, and the process denoted as gsth process is a process obtained by simplifying the GSTH process. It is assumed that only the contraction process shown in FIG. 4 is performed without performing the expansion process shown.

[Hanger detector using GSTH treatment: GSTH (A-GSTH (A))]
Details of the hanger detection apparatus using image processing according to the first embodiment of the present invention will be described with reference to FIGS.

As shown in FIG. 1, the hanger detection apparatus using image processing in this embodiment includes a line sensor camera 2 installed on the roof of a train vehicle 1 and an image processing unit 3 installed inside the vehicle 1. ing.
The trolley wire 4 is supported by a plurality of hangers 6 that are installed on the platform wire 5 at approximately equal intervals.

The line sensor camera 2 is installed on the side of the roof of the vehicle 1, the installation angle and the elevation angle so that the optical axis thereof is orthogonal to the traveling direction of the vehicle 1 and one hanger 6 is contained in the field of view. Is set. Here, an alternate long and short dash line in FIG. 1 indicates a photographing line. An image I (hereinafter referred to as a line sensor image) I as shown in FIG. 8 acquired by the line sensor camera 2 is input to the image processing unit 3.
In the example shown in FIG. 8, the trolley line 4, the overhead line 5, and the four hangers 6 are displayed in the line sensor image I.

The image processing unit 3 analyzes the line sensor image I and detects the hanger 6, and includes an arithmetic device 31 and a recording device 32.

As shown in FIG. 2, the arithmetic unit 31 includes a GSTH processing unit 31a, a hanger detection unit 31b, and a memory 31c.

The GSTH processing unit 31a performs image luminance inversion processing on the line sensor image I based on the parameters input from the recording device 32 via the memory 31c and the line sensor image I input from the line sensor camera 2 via the memory 31c. And GSTH treatment.
Here, the image luminance inversion processing is processing for inverting the luminance value for each line in the x direction with respect to the line sensor image I. However, if the luminance value of the main component in the x direction is dark, it is determined that the night or a tunnel, and luminance inversion processing is not performed.
The GSTH process is a gray scale-top hat process (Gray Scale-Top Hat process). As the name implies, the top hat process is applied to a gray scale image. The Top Hat process is a process of subtracting an opening image (an image subjected to a contraction process and an expansion process for the same number of times) from the original image.

Hereinafter, GSTH processing by the GSTH processing unit 31a will be briefly described with reference to FIGS. GSTH processing is performed according to the following flow.
(1) As shown in FIG. 4, a range of N pixels (three pixels in FIG. 3) S _D −m (m = 1 to M) is applied to the original image 11 shown in FIG. Shrinkage processing is performed by writing the luminance value of the darkest pixel to another buffer (shrinkage processing buffer) 12-m while shifting to create a shrinkage processed image 12 (12-M).
(2) Subsequently, as shown in FIG. 5, while shifting the range (hereinafter referred to as the search range) S _L -m of N pixels (three pixels in this embodiment) with respect to the contraction processed image 12, An expansion process for writing the luminance value into another buffer (expansion process buffer) 13-m is performed to create an opening image 13 (13-M).
(3) Subsequently, the opening image 13 (13-M) created in (2) is subtracted from the original image 11.
By performing the above processing, a GSTH image 14 in which bright pixels surrounded by a dark portion in the image are emphasized as shown in FIG. 6 is obtained.

After the image brightness inversion operation in GSTH processing unit 31a, GSTH image (transformed image) I _D as shown in FIG. 12 GSTH processing has been performed is input through the memory 31c to the hanger detection unit 31b.

The hanger detection unit 31b detects the hanger 6 by binarization processing or the like based on the parameters input from the recording device 32 via the memory 31c and the converted image input from the GSTH processing unit 31a via the memory 31c.
Here, since the installed hangers 6 are installed at predetermined equal intervals (distances), the position in the image can be grasped by counting the number of detected hangers 6.
Moreover, as shown in FIG. 12, only the hanger 6 installed vertically is detected in this embodiment, and the inclined hanger 6 is not detected. The number of hangers 6 between the utility poles is managed in advance as equipment data. Therefore, by separately interlocking with the utility pole detection signal, when the number of hangers 6 for each utility pole is small, it can be detected as the hangers 6 having an abnormal angle.
Information on the hanger 6 detected by the hanger detection unit 31b is output to the memory 31c as a hanger detection result.

The memory 31c temporarily stores various data. Various data are stored in the recording device 32. Note that the parameters described above are the number of pixels N for setting the search range, the number of pixels w and h required when performing GSTH processing, the binarization threshold value, and the like.

Hereinafter, the flow of processing in the hanger detection apparatus using image processing according to the present embodiment will be described with reference to FIG.

As shown in FIG. 7, in the hanger detection apparatus using image processing according to the present embodiment, a line sensor image I as shown in FIG. 8 is first input from the line sensor camera 2 (step S1). Note that the x direction shown in FIG. 8 corresponds to the photographing line of the line sensor camera 2 in FIG. Also, the y direction shown in FIG. 8 is the time taken by the line sensor camera 2 line by line. An image taken from the vehicle 1 traveling at a constant speed is as shown in FIG.

Subsequently, the GSTH processing unit 31a performs a process of inverting the luminance value for the line sensor image I (step S2). This gives inverted luminance image I _A, as shown in FIG. Note that, as described above, the luminance inversion is processed for each line in the x direction of the line sensor image I, and when the luminance value of the main component in the x direction is dark, it is determined as nighttime or tunnel and the luminance is not inverted. The tunnel portion shown in the upper part of FIG. 8 whose background is a dark color can be assimilated with a section whose background such as the sky is a light color.

Subsequently, likewise in GSTH processing unit 31a, to GSTH processing the luminance reversed image I _A shown in FIG. 8 the x direction (Step S3). That is, the contraction process shown in FIG. 4 for acquiring the darkest pixel in the region of w × 1 [pixel] long in the x direction is performed, and the expansion process shown in FIG. 5 for acquiring the brightest pixel is performed. Thus obtaining a first GSTH image I _B of white points as shown in FIG. 10 are extracted for the following width w x direction. Note that w corresponds to the width of the strip such as the trolley wire 4 and is calculated experimentally. In the example shown in FIG. 10, the first GSTH image I _B is hangers 6 and telephone poles and structures such as buildings against inverted luminance image I _A (hereinafter, a disturbance hereinafter) has a picture 7 is removed .

Subsequently, the GSTH processing unit 31a similarly subtracts the first GSTH image I _B from the luminance inverted image I _A to obtain a difference image I _C as shown in FIG. 11 (step S4). In the example illustrated in FIG. 11, the difference image I _C is an image in which the hanger 6 and the disturbance 7 such as a utility pole or a building are extracted. In addition, when the area | region which becomes negative by a difference process is made, 0 or less shall be rounded up to 0.

Subsequently, the difference image I _C is GSTH processed in the y direction in the GSTH processing unit 31a (step S5). That is, the contraction process shown in FIG. 4 for acquiring the darkest pixel in the 1 × h [pixel] region long in the y direction is performed, and then the expansion process shown in FIG. 5 for acquiring the brightest pixel is performed. As a result, as shown in FIG. 12, an image (hereinafter referred to as a second GSTH image) _{ID in} which the white portion is extracted in the y direction and with a length of h or less is obtained. Note that h corresponds to the diameter of the hanger 6 and is calculated experimentally. In the example shown in FIG. 12, the second GSTH image I _D is an image in which only the hangers 6 are extracted to the difference image I _C.
The second GSTH image _ID is input as a converted image to the hanger detector 31b via the memory 31c.

Subsequently, the hanger detection unit 31b, the second GSTH image I _D shown in FIG. 12, for detecting the hanger 6 using binarization process or the like (step S6). Since the installed hangers 6 are installed at predetermined equal intervals (distances), the number of detected hangers 6 is counted to grasp the position in the image.

The above processing from Step P1 to Step P6 is performed until the input of all the line sensor images I is completed.

In the hanger detection apparatus using image processing according to the present embodiment configured as described above, the GSTH processing is a combination of the shrinkage processing shown in FIG. 4 and the expansion processing shown in FIG. By obtaining a straight line from the 1 × h [pixel] area, it is possible to extract only a straight line that has a predetermined direction and is thinner than a predetermined thickness from the straight lines that appear in the luminance inverted image I _A.
Therefore, the hanger detection apparatus using image processing according to the present embodiment has the following advantages that make use of the features of GSTH.
-Only the hanger 6 reflected in a predetermined direction and a predetermined thickness can be detected.
-Only the hanger 6 can be detected by eliminating the disturbance 7 having characteristics different from those of the hanger 6 such as the tunnel wellhead and the utility pole 7.
In the present embodiment, only normal hangers 6 installed in the vertical direction are detected, and abnormal hangers 6 whose installation angles are outside the control value are not detected. Therefore, the abnormal hanger 6 can also be detected by comparing the number of hangers 6 detected in the present embodiment with the number of normal hangers 6 for each utility pole that is managed in advance as facility data.

[Hanger detector using simplified GSTH treatment: GSTH (gsth (A))]
Hereinafter, a hanger detection apparatus using image processing according to the second embodiment of the present invention will be described with reference to FIGS. 1 to 6 and FIGS. 8 to 13. The hanger detection apparatus using image processing according to the present embodiment is different from the first embodiment in that the processing in the GSTH processing unit 31a is simplified. The apparatus configuration is the same as that of the hanger detection apparatus based on image processing according to the first embodiment, and the description thereof will be omitted below, focusing on the processing in the image processing unit 3.

Based on FIG. 13, the flow of the process in the hanger detection apparatus by the image processing which concerns on a present Example is demonstrated.
As shown in FIG. 13, in the hanger detection apparatus using image processing according to the present embodiment, a line sensor image I as shown in FIG. 8 is first input from the line sensor camera 2 (step S11), and then In the GSTH processing unit 31a, the line sensor image I is subjected to a process of inverting the luminance value for a photographing line whose main component luminance value is light in the x direction (step S12). This gives inverted luminance image I _A, as shown in FIG.

Subsequently, likewise in GSTH processing unit 31a, to gsth processing the luminance reversed image I _A shown in FIG. 8 the x-direction (step S13). That is, the contraction process shown in FIG. 4 for acquiring the darkest pixel in the region of w × 1 [pixel] long in the x direction is performed, while the expansion process shown in FIG. 5 for acquiring the brightest pixel is not performed. As a result, a gsth image I _{C in} which white portions as shown in FIG. 11 are extracted with a width w or less in the x direction is obtained. Note that w corresponds to the width of the strip such as the trolley wire 4 and is calculated experimentally.

Subsequently, the GSTH processing unit 31a similarly performs GSTH processing on the gsth image I _C in the y direction (step S14). That is, the contraction process shown in FIG. 4 for acquiring the darkest pixel in the 1 × h [pixel] region long in the y direction is performed, and then the expansion process shown in FIG. 5 for acquiring the brightest pixel is performed. As a result, as shown in FIG. 12, a GSTH image _{ID in} which white portions are extracted in the y direction and with a length equal to or shorter than h is obtained. Note that h corresponds to the diameter of the hanger 6 and is calculated experimentally.
The GSTH image _ID is input as a converted image to the hanger detector 31b via the memory 31c.

Subsequently, the hanger detection unit 31b detects the hanger 6 from the GSTH image _ID shown in FIG. 12 using binarization processing or the like (step S15). Since the hangers 6 are installed at a predetermined interval (distance), the current position of the vehicle 1 is grasped by counting the detected hangers 6, and the installation angle of the hangers 6 has a management value. Inspect whether the installation angle is.

The above processing of Step P11 to Step P15 is performed until the input of all the line sensor images I is completed.

In the hanger detection apparatus using image processing according to the present embodiment configured as described above, the luminance inversion image I _A is subjected to gsth processing in the x direction (only the contraction processing shown in FIG. 4 is performed), thereby simplifying the processing. Therefore, there is an advantage that the hanger detection processing can be speeded up as compared with the first embodiment. Although the hanger 6 detected on the image is shortened by performing the gsth process, the present embodiment is intended to determine whether or not the hanger 6 is present. Absent.

The present invention is suitable for application to a hanger detection apparatus using image processing.

1 vehicle 2 line sensor camera 3 image processing unit 4 trolley wire 5 messenger wire 6 Hanger 7 structure 31 arithmetic unit 31a GSTH processor 31b hanger detection unit 31c memory 32 recording device I line sensor image I _A INVERSION image I _B first GSTH image I _C difference image, gsth image _ID Second GSTH image, GSTH image

Claims

A line sensor camera for photographing a hanger that is installed at equal intervals and supports a trolley line from the roof of the vehicle, and an image processing unit that is installed inside the vehicle and analyzes an image obtained by the line sensor camera, A hanger detection device by image processing for detecting a hanger in an image,
The image processing unit
A GSTH processing unit that sequentially performs GSTH processing on a line sensor image input from the line sensor camera in a shooting line direction and a time direction of the line sensor camera to obtain a GSTH image in which the hanger in the image is emphasized; ,
A hanger detection device using image processing, comprising: a hanger detection unit for detecting a hanger from the GSTH image.
The GSTH processing unit obtains a first GSTH image by performing an expansion process on the line sensor image after performing a contraction process on the imaging line direction of the line sensor camera as the GSTH process, and acquiring the first GSTH image. A difference image is obtained by taking a difference between the image and the first GSTH image, and after the shrinkage process is performed on the difference image in the time direction, the expansion process is performed to emphasize the hanger in the line sensor image Get a second GSTH image,
The hanger detection apparatus according to claim 1, wherein the hanger detection unit detects the hanger from the second GSTH image.
As the GSTH processing, the GSTH processing unit obtains a gsth image by performing only a contraction process on the line sensor image in the photographing line direction of the line sensor camera, and acquires a gsth image in the time direction on the gsth image. The GSTH image in which the hanger in the line sensor image is emphasized by performing an expansion process after performing
2. The hanger detection apparatus according to claim 1, wherein the hanger detection unit detects the hanger from the GSTH image.
The GSTH processing unit inverts the luminance value for the line sensor image, one line at a time for the shooting line direction of the line sensor camera, and the luminance value of the main component in the shooting line direction when the luminance value of the main component is light color. The hanger detection by image processing according to any one of claims 1 to 3, wherein when the value is a dark color, the GSTH process is performed after performing a luminance inversion process that does not invert the luminance value. apparatus.