JP4552409B2 - Image processing device - Google Patents

Description

  The present invention relates to a technique for photographing a pantograph and a trolley line from an ITV camera installed on a train and detecting the pantograph and the trolley line from the photographed image by image processing.

As a current technique, there is a technique that detects a pantograph and a trolley line by image processing (see Japanese Patent Application No. 2002-151840).
Here, the pantograph detection is detected by pattern matching between a video and a pantograph pattern registered in advance (see Patent Document 2).
In trolley line detection, a trolley line that appears as a straight line in an image is detected by edge detection (see Patent Document 3).
Note that Non-Patent Document 1 describes mathematical transformation used for image processing.
Japanese Patent Application Laid-Open No. 2002-139305 (Japanese Patent Application No. 2002-332841 “Pantograph Obstacle Detection Method and Device”) Japanese Patent Application Laid-Open No. 7-271979 (Japanese Patent Application No. 6-57924 "Image Recognition Method and Apparatus") Japanese Patent Application Laid-Open No. 8-14820 (Japanese Patent Application No. 6-149246 "Edge Detection Circuit for Image Processing Apparatus") "Special Feature: Mathematical Transformation Hough Transform Used in the Field of Image and Light" O plus E December 1995 pp94-pp106 "Paper: Improvement of combinatorial Hough transform considering sampling error of digital line" IEICE Transactions D-II Vol.J78-D-II No.12 pp.1767-1776 December 1995

With the current technology, when various objects appear in the image with various shapes in the image of the camera, there is a problem that these are erroneously detected.
Therefore, it is necessary to detect a pantograph or a trolley line so that there is no false detection.

An image processing apparatus according to claim 1 of the present invention that solves the above problem detects an edge of a trolley line from an input image by edge detection, votes the coordinates of the detected edge to a Hough space by Hough transformation, and is voted In the image processing apparatus that detects a trolley line that is a straight line by detecting a vertex of a Hough space, the edge detection uses a feature that the trolley line has many vertical edges on the image and has a constant thickness. The operator is added to the image to detect the edge of the trolley line, and the image is divided into a set of slits that are horizontal to the ground, and the gray value is calculated for the divided sets of horizontal slits. Gets, detecting edges interchanged operator trolley wire for daytime use and night in response to the mode of principal component analysis and histogram obtained from the obtained density values And wherein the door.

The image processing apparatus according to claim 2 of the present invention for solving the foregoing problem resides in that in Claim 1, as the origin of the Hough transform, characterized by the use of center coordinates of the detected pantograph.

The image processing apparatus according to claim 3 of the present invention for solving the foregoing problem resides in that in Claim 1, as the height of the origin of the Hough transform, characterized in that it uses the height of the detected pantograph.

The image processing apparatus according to a fourth aspect of the present invention for solving the above-mentioned problems is the image processing apparatus according to the second or third aspect , wherein the pantograph is obtained by removing an offset from coordinates of a pantograph candidate obtained by pattern matching of two or more camera images. These coordinates are collated for each camera and detected as coordinates that minimize the difference in coordinates.

An image processing apparatus according to a fifth aspect of the present invention for solving the above-described problem is the image processing apparatus according to the first, second, third, or fourth aspect , wherein when a plurality of the trolley lines are detected, a micro area centered on the highest vertex It is characterized in that it is excluded from the Hough space and the search for the next highest vertex is repeated.

An image processing apparatus according to a sixth aspect of the present invention for solving the above-mentioned problem is the first, second, third, fourth or fifth aspect , wherein when detecting a plurality of the trolley lines, the trolley lines are converted into three-dimensional coordinates. It is converted, and it is verified whether it exists in the area | region of the same level as a pantograph horizontally with the ground, The thing which does not exist in the said area | region is excluded as an erroneous trolley line.

The present invention is an apparatus for detecting the position of a pantograph and a trolley line by image processing, and has the following advantages.
1) Since the edge detection is performed by an operator focusing on the feature that the trolley line shown on the image is vertical and has a predetermined thickness, there is an advantage that the edge of the trolley line can be detected with high reliability.
2) Since the operator is switched for each slit of the image, there is an advantage that the edge of the trolley line can be detected even when the gray value of the background of the trolley line is partially different.
3) Since the origin of the Hough conversion is set to the center of the pantograph, there is an advantage that the trolley line can be detected with uniform resolution.
4) Since the origin of the Hough transform is set to the height of the pantograph, there is an advantage that the meter θ can be constrained and the calculation amount can be reduced. 5) It is obtained in advance from the position of the pantograph obtained by two or more cameras. By removing the offset, the erroneously detected pantograph is eliminated, and there is an advantage that the position of the pantograph can be detected with high reliability.
6) Since a plurality of trolley wires are simply classified by a predetermined angle θ _P and a predetermined distance ρ _p , there is an advantage that the processing is fast. Moreover, there is an advantage that the area of the vertex is surely divided even in a complicated skirt shape.
7) Using the fact that the trolley line is almost horizontal to the ground and almost the same height as the pantograph, the trolley line is detected by verifying the coordinates of the trolley line candidates of multiple cameras, so the trolley line detection is highly reliable. There is an advantage that can be.

  As described below, the best mode is a combination of the first to seventh embodiments.

  This embodiment is an image processing apparatus that detects the position of a pantograph from two or more camera images, verifies whether the position obtained by removing the offset from this position matches between the cameras, and detects the position of the pantograph. is there.

Details of this embodiment are shown in the flowchart of FIG.
First, two or more cameras 1 to n (n ≧ 2) are installed so that a pantograph is reflected, and an image is acquired (step S3).

  The pantograph area of each camera image is registered in advance as a pattern (step S1), and the relative shift of the pantograph center position between the images with reference to the camera 1 is recorded as an offset (step S2).

Next, pattern matching is performed on the images 1 to n of each camera acquired (step S4), and offsets are removed from a plurality of pantograph positions obtained for each camera (step S5).
The position where the relative difference between the positions of the images is the smallest is set as the detected pantograph position (step S6).
The above processing is repeated until the necessary image is completed (step S7).

In the present embodiment, when detecting the position of the pantograph from two or more camera images, the position of the pantograph that does not coincide with each camera is excluded, so that there is an advantage that highly reliable pantograph detection can be performed.
Note that the image processing apparatus of the present embodiment may be realized by a program that realizes the processing of the above flowchart in software.

In this embodiment, image processing that uses the feature that the trolley line has many vertical edges on the image and has a constant thickness, adds and sums the operators shown in FIG. 2 to the image, and detects only the edge of the trolley line. Device.
That is, the operator of FIG. 2 is multiplied and summed to the image, and if the calculation result is equal to or greater than the threshold value, the edge is set.
The threshold is determined experimentally.

In FIG. 2, w is set to an odd number so as to match the thickness (width) of the trolley line on the image.
Also, W ′ is set to a width of W ′ = (w + 1) / 2 so that the total sum of operators becomes zero.
Note that the accuracy can be increased by increasing the number in the vertical direction, but it is usually 1 to 3.
Nighttime operators use nighttime operators, and daytime images use daytime operators.

The signs of the night and daytime operators are reversed because the background is white and the trolley line appears black in the daytime, while the background is black and the trolley line is white due to lighting. It is for taking a picture.
The trolley line is thick at the top of the image and the trolley line is thin at the bottom of the image, and there is a change in thickness. However, this change in thickness is within the allowable value range and does not cause a problem.

This embodiment has the following advantages.
Since various objects appear in the image, there is a problem in detecting edges that are not trolley lines in normal edge detection.
In the present embodiment, the operator uses the feature that has the feature that the trolley line has many vertical edges on the image and the thickness is constant, so that it is possible to detect only the edge that is almost the same as the thickness of the trolley line. There is.

The present embodiment is an image processing apparatus that detects an edge by dividing an image into slits and replacing the daytime or nighttime operator of the second embodiment from a set of gray values inside the slit.
Details of this embodiment are shown in FIG.

In advance, as shown in FIG. 3, the image is divided into slits that are horizontal to the ground.
When the camera is tilted, it becomes an area tilted upward as shown in FIG. 3, but when the camera is installed horizontally, it becomes a slit area that is horizontal to the image.
Next, the shade values in the shaded area in the figure are acquired as a set.

The mode value of principal component analysis or histogram is obtained from this set. If this mode value is darker than the threshold value 1, the operator for nighttime and the operator for daytime are multiplied and summed in this area. If the threshold value is 2 or more, the edge is set. If the threshold value is 2 or less, 0 is set.
The threshold is determined experimentally.
In the above processing, an operator is determined for each slit in FIG. 3 to acquire an edge.
The threshold values 1 and 2 and the width in the slit y direction are determined experimentally.

  In this embodiment, the nighttime operator and the daytime operator are interchanged. Therefore, even when the background grayscale value is partially different on one image when passing through a pier at night or passing through a tunnel in daytime, There is an advantage that the edge of the trolley line can be detected.

The present embodiment is an image processing apparatus that detects the trolley line using the center coordinates of the pantograph as the origin of the Hough transformation.
This embodiment is shown in the flowchart of FIG.

First, an image is acquired (input) from the camera, and the position of the pantograph is detected from the acquired image (step T1).
When using two or more cameras, as shown in the first embodiment, the pantograph may be detected from the minimum value by removing the offset from the pattern match position.
Next, the edge of the trolley line is detected from the input image by edge detection (step T2).
At this time, the edge detection of the second and third embodiments may be used.

Subsequently, as shown in FIG. 5, the obtained trolley line edge is coordinate-transformed with the center a of the pantograph 10 as the origin (step T3).
Further, the coordinates are voted on the Hough space by the Hough transform of the equation (1) (step T4), and a straight line is detected by detecting the vertex of the value of the voted Hough space (step T5).
ρ = x cos θ + ysin θ (1)
The above processing is repeated every time an image is input (step T6).

This embodiment has the following advantages.
That is, the trolley line can be detected as a straight line, but in the normal straight line detection, the edge of the trolley line is divided or intersected by another object, and the straight line detection is difficult.
This embodiment has the first advantage that it is not easily influenced by other objects that interfere with the trolley line because it uses the Hough transform.

Also, the Hough conversion has a problem that the resolution of the straight line detected by the position of the origin becomes non-uniform.
This is because, as shown in FIG. 1 of Non-Patent Document 2, the resolution is uniform near the origin of θ = 0 and ρ = 0, but the resolution becomes non-uniform when away from θ = 0 and ρ = 0. .
On the other hand, in this embodiment, since the center coordinate of the pantograph is set as the origin of the Hough transformation, the trolley line near the pantograph is equal to the trolley line near the origin, and there is a second advantage that the trolley line can be detected with uniform resolution.

Further, in order to detect a plurality of straight lines from the Hough transform, a plurality of vertices in the Hough space are detected. In this embodiment, the shape of the vertices is almost the same on the pantograph, so that a plurality of vertices can be easily obtained. There is also a third advantage.
Note that the image processing apparatus of the present embodiment may be realized by a program that realizes the processing of the above flowchart in software.

The present embodiment is an image processing apparatus that detects the trolley line by using the height of the center coordinate of the pantograph as the y coordinate of the origin of the Hough transform.
In this embodiment, in step T3 of the flowchart shown in FIG. 4, the coordinate of the obtained trolley line edge is not converted with the center of the pantograph as the origin, but the height b of the pantograph is set as shown in FIG. The coordinates are transformed as the origin, and the coordinates are voted on the Hough space by the Hough transformation of the equation (1), and the straight line is detected by detecting the vertex of the value of the voted Hough space.
Other processes are the same as those in the fourth embodiment.

In FIG. 6, the intersection of the left edge of the image and the height of the pantograph is used as the origin, but it may be set anywhere as long as the height of the pantograph on the dotted line.
Although this embodiment has a problem that the resolution becomes non-uniform compared to the fourth embodiment, the parameter θ in the equation (1) is limited only to the angle of the trolley line that intersects the pantograph, and the calculation processing is reduced. There is an advantage of being fast.

The present embodiment is an image processing apparatus that detects a plurality of trolley lines by excluding the vertices of the Hough space obtained in the fourth and fifth embodiments with a predetermined distance θ _P and a predetermined angle ρ _p .
This embodiment is performed as follows.
First, the edge of the trolley line is detected from the input image by edge detection.
At this time, the edge detection of the second and third embodiments may be used.
The obtained trolley line edge is voted to the Hough space of FIG. 7 by Hough transformation, the highest vertex 1 is detected from the Hough space, and θ _p and ρ _p of the vertex 1 are registered as a trolley line.
The vertex of the Hough space may be obtained by the coordinate transformation in the fourth or fifth embodiment.

The hatched area surrounded by θ _p and ρ _p with the vertex 1 as the center is excluded from the Hough space as the area to which the vertex belongs, the next highest vertex is searched, and the vertex 2 is detected and excluded.
The above processing is repeated for the required number of vertices.
This embodiment has the following advantages.
That is, one trolley line corresponds to one vertex of the Hough space as shown in FIG.

In order to detect a plurality of trolley lines, it is necessary to search for vertices in the Hough space and the skirts belonging to the vertices, and to repeat this search to detect a plurality of trolley lines. .
Further, since the shape from the normal vertex to the skirt does not become a continuous profile shape such as a normal distribution as shown in FIG. 1 of Non-Patent Document 2, there is a problem that it is difficult to accurately detect the skirt by searching.

This embodiment includes a plurality of trolley wires in contact with the pantograph, areas belonging utilizing structural predetermined distance [rho _p a predetermined angle theta _P away condition of the overhead line, theta _p from the apex, the area of the [rho _p to the vertex Is simply excluded, and the next vertex is searched without searching the base.
As a result, there is a first advantage that the search for the base is unnecessary and the processing becomes fast.
In addition, there is a second advantage that the area division of the vertex is ensured even with a complicated skirt shape.

This embodiment is an image processing apparatus that verifies the coordinates of trolley lines detected from two or more camera images using 3D coordinates.
This embodiment is shown in the flowchart of FIG.
First, as shown in FIG. 10, two or more camera images in which the pantograph 10 and the trolley line 20 are reflected are acquired (step V1), an edge is detected by edge detection for each camera image (step V2), and a plurality of trolleys are obtained. Line candidates are detected (step V3).
Example 4, Example 5, and Example 6 may be used for this trolley line detection.

The coordinates of the intersection of the trolley line and the pantograph and the coordinates of the upper end of the image are respectively projected and converted into three-dimensional coordinates by stereo measurement (step V4), and as shown in FIG. It is verified whether or not the coordinates of the intersection point c of the line and the pantograph are present and the coordinates of the image upper end d of the trolley line are present, and those that are in a valid area are determined as trolley lines (step V5).
This area is an area for verifying whether the trolley line 20 is substantially horizontal and the height is the same as that of the pantograph 10, and is determined experimentally.
The above processing is repeated every time an image is input (step V6).

In this embodiment, the trolley line detected between the cameras is converted into 3D coordinates, and the upper end point of the image of the trolley line is almost horizontal with respect to the ground and the height is the same as the pantograph to eliminate the erroneous trolley line. Therefore, there is an advantage that a highly reliable trolley wire can be detected.
Note that the image processing apparatus of the present embodiment may be realized by a program that realizes the processing of the above flowchart in software.

  According to the present invention, the detected pantograph and trolley line are used for equipment maintenance depending on whether or not the detected pantograph and trolley line are reflected at a specified position.

It is a flowchart which shows the procedure of the image processing which concerns on 1st Example of this invention. It is explanatory drawing which shows the trolley line detection operator which concerns on the 2nd Example of this invention. It is explanatory drawing which shows acquisition of the mode value which concerns on the 3rd Example of this invention. It is a flowchart which shows the procedure of the image processing which concerns on the 4th, 5th Example of this invention. It is explanatory drawing which shows the origin of the coordinate transformation which concerns on the 4th Example of this invention. It is explanatory drawing which shows the origin of the coordinate transformation which concerns on the 5th Example of this invention. It is explanatory drawing which shows the area | region of the vertex of the Hough space which shows the 6th Example of this invention, and a base. It is a flowchart which shows the procedure of the image processing which concerns on the 7th Example of this invention. It is explanatory drawing which shows the three-dimensional coordinate of the trolley which concerns on the 7th Example of this invention. It is explanatory drawing which shows the two camera images which concern on the 7th Example of this invention.

Explanation of symbols

10 Pantograph 20 Trolley line a Pantograph center b Pantograph height c Intersection of trolley line and pantograph d Upper end of image of trolley line

Claims (6)

An image that detects a trolley line edge by detecting an edge of the trolley line from the input image, voting the coordinates of the detected edge to the Hough space by Hough transformation, and detecting the vertex of the voted Hough space In the processing device, the edge detection is to detect an edge of the trolley line by multiplying and adding an operator using a feature having a characteristic that the trolley line has many vertical edges on the image and a constant thickness. Yes, the image is divided into slit-like sets that are horizontal to the ground , gray values are obtained for the divided horizontal slit-like sets, and principal component analysis and histograms obtained from the obtained gray values are obtained.
An image processing apparatus for detecting an edge of a trolley line by switching a daytime / nighttime operator in accordance with a mode value of a ram . The image processing apparatus according to claim 1, wherein a center coordinate of the detected pantograph is used as an origin of the Hough transform. The image processing apparatus according to claim 1, wherein the height of the detected pantograph is used as the height of the origin of the Hough transform. The pantograph is obtained by removing an offset from the coordinates of a pantograph candidate obtained by pattern matching of two or more camera images and comparing the coordinates for each camera to detect the coordinate difference as a minimum. The image processing apparatus according to claim 2, wherein the image processing apparatus is an image processing apparatus. In the case of detecting a plurality of the trolley lines, it is possible to repeat the search for the next highest vertex by excluding the minute region centered on the highest vertex from the Hough space. 4. The image processing apparatus according to 4. When detecting a plurality of the trolley lines, convert the trolley lines into three-dimensional coordinates, verify whether they exist in an area that is horizontal to the ground and approximately the same height as the pantograph, and do not exist in the areas 6. The image processing apparatus according to claim 1, wherein the image processing apparatus is excluded as an erroneous trolley line.

Images