JPH08287232A - Method for removing unmonitored object for image processing - Google Patents

Abstract

PURPOSE: To improve detection accuracy without narrowing down a monitor area by finding and setting an unmonitored area where an unmonitored object is possibly present in advance by learning within the minimum limit. CONSTITUTION: The difference between an input image 2 of, for example, a Chinese character in the Figure as an unmonitored object and a reference image 1 of the Chinese character is calculated to obtain plural difference images 3. Then the centers of gravity of the respective difference images 3 are found to obtain a difference gravity center group image 4 and an area where the unmonitored object is possibly present, i.e., an unmonitored area 5 is found. After the learning in this way, a monitored object 6 is found first to inspect the actual monitored object, and the difference image 7 between this object and reference image 1 is found. The difference image 7 include a detected object, so respective difference images 7 in one screen 1 are labeled and partitioned. The centers of gravity of the respective bodies are found from the difference images 7 and when the coordinates of a gravity center pixel are in the unmonitored area 5, the labeled image is determined as a difference labeling image 8, which is deleted to obtain a result 9.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing method capable of reliably removing a disturbance (non-monitoring target) other than a monitoring target.

[0002]

2. Description of the Related Art When a liquid leak such as oil is inspected by image processing, if there is a disturbance such as a "flip" that shakes due to wind or some vibration in the field of view to be monitored, conventionally, for example, There is a method in which a portion of the image monitoring area where disturbance may exist is painted black in a rectangular area (ROI or the like) and ignored so that the change in this portion is not seen.

[0003]

However, in the method using the rectangular screen as described above, there is a tendency to set a wide range in consideration of some safety from the viewpoint of preventing an error, so that it becomes a detection target (monitoring). There is a problem that the area becomes narrow.
Therefore, an object of the present invention is to improve the detection accuracy without narrowing the monitoring area.

[0004]

In order to solve such a problem, according to the invention of claim 1, the process of obtaining the center of gravity of the image of the difference between the input image of the non-monitored object and its reference image is the input. By changing the image and sequentially executing it, the area that may exist in the monitoring screen of the non-monitoring object is determined in advance as a non-monitoring area, and then the image and the non-monitoring object of the monitoring object are determined. The center of gravity of the image of the difference from the reference image is obtained, and when the center of gravity is in the non-monitoring area, it is determined that the object is not a monitoring target and that image is not adopted.

According to the second aspect of the present invention, a rectangular image area which is slightly larger than the portion of the unmonitored object is set in advance, and the image of the monitored object and the reference image of the unmonitored object are set. The difference is binarized, a label number is attached to the obtained difference binarized image, and a logical product operation of this differential binarized image and the rectangular image area is performed to obtain a center of gravity and brightness of the logical product image. The differential binarized image having the label number whose luminance matches the luminance of the differential binarized image to which the label number is attached is not an object to be monitored, and the image is not adopted.

[0006]

The non-monitored area in which the non-monitored object may exist is not set as a rough area such as a rectangular area but is set in advance by learning with a minimum size, thereby setting the monitored area. The detection accuracy is improved without narrowing. Further, by setting the non-monitoring region as a region that is slightly larger than the non-monitoring target, the monitoring region is not made narrower than in the past, and the complexity of learning is eliminated.

[0007]

1 is a functional block diagram showing a first embodiment of the present invention. This allows ITV to be inspected or monitored
The procedure and the image obtained as a result of imaging with an imaging means, such as a camera, and performing with an image processing apparatus are shown. First, learn about "tags" as non-monitored objects. That is, an image of the "tag" is picked up by an image pickup means (not shown) to obtain the input image 2. Next, a difference image 3 is obtained by taking the difference between the input image 2 and the reference image 1 of the “tag”. Note that a plurality of difference images 3 can be obtained by sequentially changing the input image 2 to capture an image and obtaining the difference image 3 thereof.

Next, the center of gravity of each difference image 3 is obtained to obtain the difference center of gravity group image 4, and by including all of these center of gravity, an area in which an unmonitored object may exist,
That is, the non-monitoring area 5 is obtained. At this time, the non-monitoring area 5
Is quantized data "0" and other areas are "25".
5 ”will be assigned.

After the learning in this way, when inspecting an actual monitored object, first, the monitored object image 6 is obtained, and the difference image 7 between this and the reference image 1 is obtained. Since an object to be detected is also included in this difference image, for example, a label is attached to each of the difference images in one screen (labeling) to distinguish them.

Then, the center of gravity (Gx, Gy) of each object is obtained from the difference image 7, and the center of gravity (Gx, Gy) is calculated.
y) It is checked whether the coordinates of the pixel are in the non-monitoring area 5, and if it is in the non-monitoring area 5, the labeled image is determined as the differential labeling image 8 and the image is deleted to obtain the result 9.

FIG. 2 is a functional block diagram showing a second embodiment of the present invention. In this example, first, a rectangular image area that is slightly larger is set in a portion such as a bill. This is indicated by a dotted line in FIG. 2 as a setting area (non-monitoring area) 15. Next, a difference image 13 is created from the target image 12 and the reference image 11 of the bill, and after binarization, labeling is performed to obtain a label image 14.

On the other hand, the logical product of the difference image 13 and the setting area 15 is obtained to obtain the logical product image 16 and its center of gravity (Gx, G
y) and the luminance at that position are obtained as indicated by reference numeral 17. For the label image 14, for each number (here, 1 to 4)
Since the center of gravity (Gx, Gy) and the brightness at that position are obtained, the object (label number) that matches the brightness of the logical product image 16 is obtained as indicated by reference numeral 18 and is deleted from the label image 14. The result 19 is obtained.

In the above, the method of removing the "swing" phenomenon of the "tag" in the oil leak detection of the plant has been mainly described. It is also applicable to the case of monitoring the site abnormality with the camera of (1), the site abnormality monitoring with the inspection robot in the sewage plant, and the case of monitoring the image of the instruments installed beside the expressway.

[0014]

According to the present invention, the non-monitoring area is not determined roughly as in the prior art, but is determined as a minimum area by learning. Therefore, the object can be detected without narrowing the monitoring area. Therefore, there is an advantage that the detection accuracy can be improved. Also, by setting the non-monitoring area as a rectangular image area that is one size larger than the non-monitoring object, it is possible to eliminate the troublesome processing due to learning, but to make the area smaller than the conventional one, and to make the monitoring area as narrow as possible. You can

[Brief description of drawings]

FIG. 1 is a functional block diagram showing a first embodiment of the present invention.

FIG. 2 is a functional block diagram showing a second embodiment of the present invention.

[Explanation of symbols]

1, 11 ... Reference image, 2 ... Input image, 3, 7, 13 ... Difference image, 4 ... Centroid group image, 5 ... Non-monitoring region, 6, 12 ...
Target image, 8, 14 ... Label image, 9, 19 ... Result, 1
5 ... setting area, 16 ... AND image, 17 ... barycentric position, 1
8 ... Brightness matching image.

Continued front page (72) Inventor Osamu Takada 1-1, Tanabe Nitta, Kawasaki-ku, Kawasaki-shi, Kanagawa Fuji Electric Co., Ltd.

Claims (2)

[Claims] 1. A process for obtaining the center of gravity of an image of the difference between an input image of a non-monitoring target and its reference image is sequentially executed by changing the input image, thereby displaying a screen within the monitoring screen of the non-monitoring target. A region that may exist is determined in advance as a non-monitoring region, and thereafter, the center of gravity of the image of the difference between the image of the monitoring target and the reference image of the non-monitoring target is obtained, and the center of gravity is the non-monitoring region. A method for removing a non-monitored object in image processing, characterized in that the image is not adopted when it is inside the object. 2. A rectangular image area which is slightly larger than the portion of the non-monitored object is preset, and the difference between the image of the monitored object and the reference image of the non-monitored object is binarized. , A label number is attached to the obtained differential binarized image, and a logical product operation of this differential binarized image and the rectangular image area is performed to obtain a center of gravity and luminance of the logical product image, and the luminance is Difference 2 with label number
A method of removing a non-monitoring target in image processing, characterized in that a difference binarized image of a label number that matches the brightness of a binarized image is not a monitoring target, and that image is not adopted.