JP2001266271A - Method for automatically detecting generation of dust and smoke - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method for automatically monitoring the generation of dust and smoke capable of detecting the generation of dust and smoke even in the case the positional deviation of a monitored image takes place due to the vibration, etc., of an image pickup device for a monitoring image. SOLUTION: First, when a generated dust and smoke detection device is started, a starting signal is simultaneously applied to the CPU of a variable density image processing board. When the variable density image processing board is started, the variable density image processing board outputs an instruction to input an image signal from the image pickup device to a color image board and also inputs the image signal distributed by an image pickup device controller to the self-board. Then, the variable density image processing board input an image and performs shift variable detection processing with respect to a variable density image inputted to the self-board the last time. Namely, the two-dimensional cross-correlation between the input image and the image inputted the last time is taken, and it is decided that the images coincide at a position where a cross-correlation value is maximum. Thereafter, a part where the degree of variable density changes equally to or more than a threshold is detected by comparing the images. This becomes a dust and smoke generation candidate from the variable density image.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for automatically detecting dust and smoke using an imaging device.

[0002]

2. Description of the Related Art At present, in companies that have factories that can generate dust and smoke, such as steel mills and chemical factories, it is essential to monitor the state of dust and smoke from these factories for environmental protection. Work. While many companies still carry out this work by visual observation, there have been many reports of imaging places where dust and smoke are expected and detecting these dust and smoke using image processing. As a detection method by image processing, for example, JP-A-4-263394,
JP-A-5-12407, JP-A-8-315113
There is a technique disclosed in Japanese Unexamined Patent Publication (Kokai) No. H10-26095.

[0003] JP-A-4-263394, JP-A-5
The technology disclosed in JP-A-12407 discloses imaging an imaging target at different points in time and detecting dust and smoke based on temporally changed image information (pixels).
The technique disclosed in Japanese Patent Application Laid-Open No. 8-315113 does not use a pixel as a unit of comparison but divides a monitoring area into a plurality of blocks and introduces a macro concept such as calculating an average luminance for each block. It is a detection method. However, this method also detects dust and smoke based on the temporal change of the average luminance of each block. These detection methods are based on the premise that the installed CCD camera always captures the same position at the same magnification without being disturbed by the size regardless of the imaging time.

[0004]

However, these CCD cameras for monitoring are controlled by a pan head capable of high-precision positioning, or even if they are completely fixed, they are used for monitoring dust and smoke. In general, since the camera is installed outdoors and at a high place, a slight image shift occurs due to wind or ground vibration.

[0005] For this reason, a method of detecting a temporal change for each pixel and detecting dust / smoke by using the method disclosed in JP-A-4-263394 and JP-A-5-12407. , There is a problem that it is difficult to perform stable detection because the same pixel is imaged at different times depending on the time.

The technique described in Japanese Patent Application Laid-Open No. 8-315113 performs a detection method based on the average of each block.
As compared with the technique described in Japanese Patent Application Laid-Open No. 12407, the effect of pixel shift is reduced by the amount of macroscopic detection. However, when it is assumed that there is a pixel shift, a process of detecting a change must ignore a small change to some extent as an effect of the pixel shift.

On the other hand, the state of generation of dust and smoke to be detected is random, and dust and smoke which are difficult to detect even by visual observation are imaged only very thinly. Therefore, in order to actually detect dust and smoke, it is necessary to detect a minute change in the image. Therefore, Japanese Patent Application Laid-Open
Even if the technique described in JP-A-315113 is used, there is a problem that it is difficult to detect random dust and smoke or dust and smoke that are difficult to detect even visually.

[0008] The present invention has been made in view of such circumstances, and even when the position of an image to be monitored is displaced due to vibration of a monitoring image pickup device or the like, random dust / smoke or visual observation is possible. An object of the present invention is to provide an automatic dust / smoke monitoring method capable of detecting dust / smoke which is difficult to detect.

[0009]

According to a first aspect of the present invention, a space area where dust or smoke is likely to be generated is captured by an imaging device, and a time series of the obtained image is obtained. A method of automatically detecting dust and smoke from changes,
Automatic detection of dust / smoke emission, characterized by performing pattern matching on images to be compared and matching pixels corresponding to the same field of view before obtaining a time-series change of the obtained image. A method (claim 1).

In this means, prior to comparing the images taken at a certain time interval, pattern matching of the compared images is performed to make the pixels corresponding to the same field of view correspond. As a method of pattern matching, a known method such as a method using cross-correlation can be used. Therefore, since the compared pixels correspond exactly to the same visual field, even if the imaging device causes an imaging shift due to wind or ground vibration, the pixels of different visual fields are compared and an erroneous determination is made. No more falling. Therefore, a minute change in a minute field of view can be detected.

The "time-series change of an image" includes the amount of change in luminance and color of each pixel from a steady value, the rate of change in luminance and color of each pixel over time, and the number of pixels in a block of an image. Of the average value of the brightness and color of the image from the steady value, the temporal change rate of the average value of the brightness and color of the pixels in the block of the image, and the like.

A second means for solving the above-mentioned problem is as follows.
The first means, wherein a plurality of images are taken by scanning a field of view by an imaging device, and a plurality of obtained images are connected to obtain one image, and the obtained one image A method of automatically detecting dust and smoke from a time-series change, wherein a plurality of obtained images are connected to obtain one image, and overlapping image portions are connected by pattern matching. (Claim 2).

In this means, the field of view is scanned by the image pickup device to perform image pickup a plurality of times, and a plurality of images partially overlapping each other are obtained. Then, by performing pattern matching on a portion of the overlapping image, a plurality of images are connected to obtain one image, and the obtained image is compared with a past image to obtain a time-series change of the image. ing. Therefore, even when a wide field of view is monitored by one camera, it is possible to compare pixels corresponding to the same field of view.

A third means for solving the above-mentioned problem is as follows.
In the first means or the second means, when a time-series change of an obtained image exceeds a threshold, a time during which the threshold is exceeded is a predetermined time t during a past predetermined time T. An alarm is issued in the above case, and in other cases, no alarm is issued and only the occurrence is recorded (claim 3).

In the conventional method for detecting dust and smoke,
When the chronological change of the obtained image exceeded the threshold value, a warning of dust and smoke was issued immediately. However, some dust and smoke are emitted only for a very short period of time, and such actions need not be taken immediately if a record is kept. In this means, an alarm is issued when the time exceeding the threshold value is equal to or more than the predetermined time t during the past predetermined time T. That is, the past T
During the period of time, dust is generated intermittently or continuously for the time t.
It only alerts when smoke is detected, prompting you to take action. Therefore, it is possible to prevent an unnecessary alarm from being issued.

A fourth means for solving the above-mentioned problem is as follows.
The first means has a small second threshold value different from the threshold value (first threshold value), and issues a warning when a time-series change of an obtained image exceeds a second threshold value. When a warning and a warning are issued, the value of the first threshold value is automatically changed by inputting the appropriateness (claim 4).

In this means, a human recognizes that a warning and a warning have been issued, and inputs the appropriateness. Thereby, the threshold value for issuing an alarm is automatically changed. For example, when it is determined that there is no problem in the amount of generated dust and smoke when an alarm is issued, when the fact is input, the first threshold is increased. Further, when it is determined that the warning is at a level at which a warning is originally issued when the warning is issued, the first threshold value is reduced by inputting that fact.

A fifth means for solving the above problem is as follows.
The third means or the fourth means, wherein a time-series change of the obtained image is graphed and recorded, and a person who observes the graph changes the threshold value, the predetermined time T, and the value of the predetermined time t. Can be changed (claim 5).

In this means, since the human can observe the graph and change the threshold value, the predetermined time T, and the predetermined time t, it is possible to optimize the level at which an alarm is issued by learning. Become.

A sixth means for solving the above-mentioned problem is as follows.
Any one of the first to fifth means, wherein a ratio between the intensity of a specific color in the image and the intensity of another specific color in the image is used as a time-series change of the image. (Claim 6).

In general, dust and smoke generated from a monitored place have a color unique to the place. For example,
The dust generated from converters in steel works is often brownish. Therefore, by using the ratio of the intensity of a specific color in an image to the intensity of another specific color as a time-series change of an image, it is more accurate than using a time-series change in luminance as in the related art. Dust and smoke can be detected.

[0022]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 shows Embodiment 1 of the present invention.
1 is a block diagram illustrating an overall configuration of an automatic dust and smoke monitoring device for implementing an example of a dust and smoke automatic monitoring method. In the figure, 1 is an imaging device, 2 is an electric pan head, 3 is an imaging device control device, 4 is a dust and smoke detection device, 5 is a dust detection device BUS, 6 is a grayscale image processing board, 7 is a color image board, 8 is a CPU board, 9 is a VGA board, 10
Is a PIO board, 11 is a LAN board, 12 is a CRT,
Reference numeral 13 denotes a status indicator lamp, and reference numeral 14 denotes a speaker.

The image pickup device 1 is fixed to the electric pan head 2 and is installed at a high place where a monitoring area can be desired to monitor dust and smoke. The imaging device control device 3 drives the electric pan head 2 and supplies an image captured by the imaging device 1 to the dust / smoke detection device 4. The dust emission / smoke detection device 4 includes a CPU board 8 that performs overall control processing, and a VGA that performs monitoring control for displaying a monitoring state on the CRT 12.
Board 9 and LA for reporting monitoring results to external devices
N board 11, a PIO board 10 for controlling an indicator light 13 and a speaker 14 for notifying an alarm when dust or smoke is detected
And a grayscale image board 6 for performing main image processing of the present invention.
And a color image board 7.

In this embodiment, the difference between the grayscale image board and the color image board is that the processing for increasing the calculation capacity such as the displacement correction as described later is performed on the grayscale image to generate dust and smoke. The calculation such as detection is performed using a color image.

The imaging device controller 3 distributes the image signal and supplies it to the grayscale image processing board 6 and the color image board 7. The gradation image processing board 6 is equipped with a CPU capable of operating asynchronously with the CPU of the dust / smoke detection device 4 and a processor dedicated to image processing, and always measures the displacement of the captured image separately from the monitoring cycle. Going fast.
The color image board 7 inputs a color image in a monitoring cycle and performs dust emission / smoke detection and determination based on information from the density image processing board 6.

Next, the processing procedure in the grayscale image processing board 6 will be described with reference to FIG. First, an activation signal is applied to the CPU of the grayscale image processing board when the dust / smoke detection device 4 is activated. Upon activation, the imaging device control device 3
Is input to the own board.

Subsequently, the gray-scale image processing board 6 receives the image and performs a process of detecting the amount of positional deviation from the gray-scale image previously input to the own board. That is, a two-dimensional cross-correlation between the input image and the previously input image is obtained, and it is determined that the images match at a position where the cross-correlation value is maximum. After that, a comparison is made between the images to detect a portion where the gray level has changed by a threshold value or more. This is a dust / smoke candidate from the grayscale image.

Next, the entire process of detecting dust and smoke will be described with reference to FIG. First, an image at time t = T from the imaging device 1 is input to the color image board 7 in a monitoring cycle of the dust emission / smoke detection device 4 (a cycle of a predetermined multiple of the processing cycle of the grayscale image processing). Subsequently, position shift correction is performed based on the position shift amount (time t = T and t = T-1 image shift amount) detected from the grayscale image processing board.

If the process up to the displacement correction is performed using a color image, an enormous amount of calculation is required. In this embodiment,
The calculation for correcting the position shift is performed using the grayscale image, and the position shift of the color image is corrected based on the result, so that the calculation amount and the calculation time are greatly reduced.

In order to align the data of the grayscale image board and the color image board, if a signal read from the image sensor at the same timing is written to both boards at the same timing, a positional shift usually occurs. There is no need for it. Therefore, the displacement amount determined based on the grayscale image data may be used as it is as the displacement amount of the color image information. However, if there is a possibility that a displacement may occur between the two depending on the writing method, the RBG information of the color image board is expressed as (R + G + B) /
After conversion into brightness information as No. 3, position alignment may be performed by a known method such as cross-correlation with data on a grayscale image board. In this case, the displacement amount of the color image is corrected by adding the displacement amount between the gradation image board and the color image board to the displacement amount determined based on the gradation image data.

A change in color is calculated for each pixel between the images whose position has been corrected, a portion where the color has changed by a threshold value or more is extracted, and dust and smoke generation candidates are extracted. To calculate the change in color, it is common to calculate the ratio or difference between a particular color and another particular color. Which color is used for the calculation is determined according to the expected color of dust and smoke.

At this time, the information on the extraction of dust and smoke candidates by the gradation image processing is compared, and noise information is removed from the dust and smoke candidates. That is, among the portions where the grayscale image has changed to the threshold value or more, those which are clearly considered to be not smoke or dust are removed. For example, an object having a large change in shading only in a very small area is an object such as an airplane or an object reflecting sunlight, and is therefore excluded from the candidate areas.

Next, the detection data of the dust emission / smoke emission candidate is plotted on a graph together with the results of the past plural times. The following method can be adopted as a method of graphing a time-series change of an image.

First, pixels whose change amounts of respective pixels exceed a preset range are extracted between the images after the registration based on the grayscale image. Then, in the extracted image, a region in which a region formed by adjoining neighboring pixels has an area equal to or larger than a predetermined area is extracted. Then, the area of this region and the average value of the luminance change amount of each pixel are plotted as graph values.

Further, a portion corresponding to the extracted area is extracted from the color image, and the chromaticity, the hue, the average value of the saturation, and the average values of the R, B, and G signals of the extracted portion are plotted as graph values. I do.

In addition, these values are calculated, and values having a strong correlation with dust and smoke generated at the place are plotted as graph values. As an example of this, if you want to focus on red-brown dust that is likely to occur in steelworks,
Plot R / B values.

The alarm output based on dust / smoke detection is performed when the value of this graph exceeds a threshold. As will be described later, an alarm is output in consideration of past data.

At the actual monitoring site, when a dust or smoke is generated, a countermeasure can be taken if the work must be performed immediately at the site where the dust or smoke is generated, or the fact that the occurrence has been recorded as data. There is no state. However, these two states cannot be divided by a single threshold. According to the present invention, in a gray zone in which judgment cannot be made based on past data, a warning is divided into ranks as cautions, and a visual judgment of a person is prompted. Based on this, there is provided a learning function for updating the determination level when the same state occurs next time. As a result, it is possible to eliminate the oversight of dust and smoke caused by a mistake in setting the threshold value and to prevent generation of a large amount of false information, thereby enabling stable automatic monitoring.

Next, a method for judging the state of generation of dust and smoke will be described with reference to FIG. (A) is an ideal graph with no generation of dust and smoke from time T0 to T. (B)
The case where there is a possibility that dust and smoke are gradually generated from T1 to T, and the case (c) is the case where there is a possibility that dust and smoke are generated rapidly. At this time, (w1-W) and (w2-W) are conventionally used criteria, but in the present invention, (w2-W) / (T
The determination is made based on the shape of the graph, such as the amount of change per unit time or the integrated amount of change per unit time as in 2-T).

(D) is a state in which dust and smoke are frequently generated, but a large amount is generated at first, but then a small amount is generated intermittently. (E) is a state in which a large amount is generated intermittently. This is a state where dust and smoke are generated. Also in this case (d)
As described above, once a large amount occurred in A, but then a small amount occurred. Therefore, it is possible to accumulate data without immediately issuing an alarm output and to issue a warning (cautionary output) to check the situation. I have. In (e), each dust / smoke A is intermittent and does not result in an alarm output, but an alarm is output on condition that there is a plurality of dust / smoke generations per unit time.

As an example of this determination, when the threshold value is exceeded at time T, it is determined whether there is time exceeding the threshold value by going back to the past up to T ₀ and determining whether the time exceeds the predetermined time t. A method is possible in which an alarm is issued when the number exceeds the limit. And the value of (T−T ₀ ) and t is
It is preferable that a human can see and change the shape of the graph.

FIG. 5 is a view for explaining an example of automatic monitoring in the image pickup apparatus 1 which is automatically panning by utilizing the position correction function according to the present invention. In (a), W is an area to be monitored, and Wn represents a range of one field of view of the imaging apparatus.
Conventionally, when the range of W is set as the monitoring area, a plurality of imaging devices 1 are installed to secure the range of W, or the motorized pan head 2 and the imaging device control device 3 use a facility having an accurate positioning mechanism to set the specified position. The imaging device 1 must be moved to a state where a plurality of the imaging devices 1 are installed in a simulated manner.

However, by using the image position correction function of the present invention, as shown in FIG. 3B, the image G1 → G2 → G3 → G4 →
... → G (n−1) → Gn, first, a detection process is performed on the same area A1 in both images by correcting the image positions of G1 and G2. Next, based on the image position correction of G3 and G4, detection processing is performed using the same area A3 in both images. In this way, by repeating the processing up to An-1, monitoring can be performed by the image capturing apparatus 1 that is auto-panning the monitoring area W. In addition, depending on the speed of the auto pan, if the processing for G2 and G3 is performed with the capability of the dust / smoke detection device 4,
More accurate monitoring can be performed.

FIG. 6 is a diagram showing a change in the average value of the color (rbg) signal of a specific block in an image when dust occurs. The horizontal axis in FIG. 6 indicates the number of times of data collection. In this case, data is taken at intervals of 2 seconds, and a value obtained by doubling the number on the horizontal axis corresponds to the number of seconds. The vertical axis is R,
The relative amounts are the amounts of change in the B and G signals. Since the dust is reddish brown, the intensity of the r signal is high and the level of the b signal is low. Therefore, it can be understood that dust can be detected by calculating the ratio of the r signal and the b signal.

[0045]

As described above, if the method for detecting dust and smoke according to the present invention is used, it is possible to remove unevenness in imaging caused by various factors, and conventionally, it is possible to detect the unevenness of image by hiding behind the unevenness in imaging. It has become possible to detect minute dust and smoke that did not exist. Further, by having a position correction function for each field of view, it is possible to automatically monitor an image capturing apparatus that is auto-panning. Further, the generation of dust and smoke can be accurately detected by the learning process.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an overall configuration of a dust / smoke detection device for carrying out the present invention.

FIG. 2 is a flowchart illustrating an example of an imaging position correction process in the dust emission / smoke detection device.

FIG. 3 is a flowchart showing the entire process of detecting dust and smoke in the dust and smoke detection device.

FIG. 4 is an explanatory diagram showing a dust / smoke detection determination method in the dust / smoke detection device.

FIG. 5 is an explanatory diagram showing an embodiment of an auto-pan imaging device in the dust emission / smoke detection device.

FIG. 6 is a diagram illustrating an example of a level change of an RGB signal when dust occurs.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Image pickup device 2 Electric pan head 3 Image pickup device control device 4 Dust and smoke detection device 5 Dust detection device BUS 6 Gray image processing board 7 Color image board 8 CPU board 9 VGA board 10 PIO board 11 LAN board 12 CRT 13 State Display, etc. 14 Speaker

 ────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Souichi Yano 7-1-1, Higashi-Narashino, Narashino-shi, Chiba F-term in Hitachi Keiyo Engineering Co., Ltd. 5C054 AA01 CA04 CC02 CH01 EA01 EA05 EA07 EE00 FB03 FC00 FC12 FC16 FD00 FE11 FE17 FE28 FF03 FF06 GB01 GD01 HA18 HA20 5C087 AA02 AA03 AA08 AA19 AA32 AA44 DD03 DD04 DD20 EE08 GG02 GG19 GG30 GG31

Claims (6)

[Claims] 1. A method for automatically detecting dust and smoke from a time-series change of an obtained image by capturing an image of a space area where dust or smoke may occur by an image pickup device. Prior to obtaining a time-series change of the obtained image, performing a process of performing pattern matching of images to be compared to correspond pixels corresponding to the same field of view, and automatically detecting dust and smoke. . 2. The method for automatically detecting dust and smoke according to claim 1, wherein a field of view is scanned by an image pickup device, a plurality of images are taken, and a plurality of obtained images are joined to form one image. In the method of obtaining dust images and automatically detecting dust and smoke from the time series change of one of the obtained images, when obtaining one image by joining the obtained plurality of images, The automatic detection method of dust and smoke, characterized in that the parts are connected by pattern matching. 3. The method for automatically detecting dust and smoke according to claim 1 or 2, wherein when a time-series change of an obtained image exceeds a threshold, the change exceeds the threshold. Automatic detection of dust and smoke characterized by issuing an alarm when the time is equal to or more than a predetermined time t during the past predetermined time T, and in other cases, not generating an alarm and only recording the occurrence. Method. 4. The method for automatically detecting dust and smoke according to claim 3, wherein the method has a small second threshold different from the threshold (first threshold), and a time series of the obtained image. A warning is issued when a specific change exceeds a second threshold value, and when a warning and a warning are issued, whether or not the warning and the warning are appropriate is input.
Automatic detection method of dust and smoke, characterized by automatically changing the threshold value of. 5. A method for automatically detecting dust and smoke according to claim 3 or 4, wherein a time-series change of an obtained image is graphed and recorded, and a person who observes the change is graphed. Wherein the threshold value, the predetermined time T, and the predetermined time t can be changed. 6. One of claims 1 to 5
The method for automatically detecting dust and smoke described in the paragraph, wherein the ratio of the intensity of a specific color in the image to the intensity of another specific color in the image is used as a time-series change of the image. -Automatic detection of smoke.