JPH0337357B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to an image monitoring system that issues an alarm by detecting a change in a monitoring image by a TV camera.

[Background Art] The present inventors have already proposed a method for detecting changes by projecting a difference image between the current image and a reference image, compressing the image information of the changed area, and transmitting it in a narrow band. However, a reliable method for identifying intruders, etc., had not yet been completed.

[Object of the Invention] The present invention was made in view of the above-mentioned problems, and its purpose is to provide an image monitoring method that can reliably detect changes caused by intruders or abnormalities such as fire. There is something to do.

[Disclosure of invention] The present invention will be explained below with reference to Examples.

Figure 1 shows the overall configuration of this embodiment.
The monitoring area is imaged by a TV camera 1 that uses an imaging device such as a CCD that is sensitive to infrared regions, and the TV camera 1 is attached to the ceiling of the monitoring area as shown in FIG. Is the image captured by this TV camera 1 in the infrared region?
The visible light range can be selected by switching the band filter 32. The A/D converter 2 is for digitizing the video signal from the TV camera 1, and the digitized image data is stored in a frame buffer. This image data capture is called "memory capture" and is usually set to about 0.5 seconds for normal intruder detection, although it depends on the moving speed of the object for change detection.

In the embodiment, there are three frame buffers, the current image frame buffer 3 is for storing the latest image data, and the reference image frame buffer 5 is for storing background image data of a monitoring area where there are no intruders etc. It is for remembering. The previous image frame buffer 4 is for storing the previous captured image.

By the way, there are two types of image change detection: comparing the current image with a background reference image, or comparing the current image with the previous image. In the former, even if the intruder is stationary, it is considered that there is a change, and the intruder is detected. However, in the latter case, when the intruder moves, only the part that moves is considered as the changed area, and the position of the intruder in the previous image and the position of the intruder in the current image are different. If so, the position of the intruder in the previous image becomes a disappearing change area in the current image, and the position of the intruder in the current image becomes an appearance change area. Therefore, when comparing the background reference image and the current image, only the appearance change area is included.

Embodiments of the present invention mainly perform abnormal change detection by comparing the current image with a background reference image, and either perform image comparison normalized for illuminance changes, or perform image comparison again when it seems that the illuminance has changed. This relates to an image monitoring method that updates the reference image of the background without regarding changes in illuminance as abnormal changes, but it also detects signs of illuminance changes when comparing the previous image and the current image, and then updates the image again. Of course, it is also possible to perform comparison and normalization with respect to illuminance.

In this embodiment, three types of methods for detecting signs of a change in illuminance are employed, and the method can be switched by an illuminance change detection method setting circuit 6.

First, to give an overview of the three methods, the first method calculates the change in the output of the illuminance meter 7 using the illuminance change calculation circuit 8, and the second method calculates the magnitude of the change for each pixel. This is a method in which a histogram is created by a histogram creation circuit 9, and a third
The method uses a continuous change determination circuit 10 that determines continuous changes in a plurality of pixels.

The method using the illumination meter 7 detects the amount of change (deviation) between the illuminance detected by the illumination meter 7 at the time of capturing the current image and the illuminance at the time of capturing the reference image (or previous image) (or the illuminance of the previous image with respect to the reference image). ) is calculated by the illuminance change calculation circuit 8. At this time, if the amount of illuminance change (deviation) is larger than a certain set value, it is not considered as an abnormal change, and the reference image update circuit 24
The reference image is updated with the current image. Also, from the calculated value of the illuminance change calculation circuit 8, a calculation circuit that calculates a correction coefficient (function) for correcting each pixel of the reference image so that it has the same conditions as the illuminance of the current image, and calculates the magnitude of change for each pixel. 26, the correction coefficient (function) calculation circuit 11 calculates the correction coefficient (function) for each pixel, and registers the calculated correction coefficient (function) in the correction coefficient (function) buffer 12. And illuminance normalization calculation circuit 1
3, by multiplying each pixel of the reference image by the product of the correction coefficient registered above and the amount of change calculated above, or by multiplying each pixel of the reference image by a function, the illuminance can be determined in the comparison between the current image and the reference image. Perform normalization for . Of course, the correction amount may be decreased or added.

Here, when the auto iris circuit or automatic gain adjustment circuit of TV camera 1 is working, the correction coefficient is such that the average brightness of the entire image is kept almost constant, so the portion that becomes a shadow when the illuminance increases is a negative correction coefficient. In other words, after making the average value the same, the brightness of each pixel of the image is corrected. On the other hand, in the case of an image that is not affected by the auto iris circuit or automatic gain adjustment circuit, the correction coefficient is
Corresponding to the reflectance of light in the direction of the TV camera 1, for example, after the correction calculation is performed by the correction calculation unit 13a of the illuminance normalization calculation circuit 13 configured as shown in FIG. 2, the gain adjustment is performed by the gain adjustment unit 13b. It is possible to simulate an auto iris circuit or an automatic gain adjustment circuit.

Furthermore, if the lighting conditions of the monitoring area change, the shadow area will also change, so a correction coefficient (function) must be found each time, but if the lighting conditions are constant, it is necessary to This will prevent errors due to influence.

Next, detection of signs of illuminance changes using a histogram will be described. First, the histogram of change for each image is a histogram (frequency distribution) of the magnitude of change for each pixel of the difference image obtained by calculating the absolute value of the difference between the current image and the reference image or previous image for each pixel of the change detection grid. , a histogram of the magnitude of change for each pixel is created by the histogram creation circuit 9. If there is a change in illuminance, it will change throughout the image, so the proportion (or number) of pixels (zero pixels) with a change smaller than a certain setting value will be small, so if we test this condition, This allows you to skip the change detection process with the next image and capture. This change detection skip is performed by the reference image update determination circuit 25. Of course, the histogram method can be included in the test of the magnitude of change for each pixel to obtain the projection described later, and when the proportion of zero pixels is small, it can be used as an illuminance change and transferred to the next change detection process. .

Now, in the example of FIG. 1, if normalization of illumination changes is not performed, the illumination off image as shown in the photograph of FIG. 14 is used as the current image, and the illumination off image as shown in the photograph of FIG. 15 (FIG. 6) In an example where the on image is used as the reference image, the histogram of the current image (as shown in Figure 16)
When compared with the histogram of the reference (background) image (shown in FIG. 18), it can be seen that there is a large difference in the average value. Here, the average value of the histogram of the current image is 42, and the average value of the histogram of the reference image is
It is 142. The histogram of the difference image in this case is as shown in FIG. Here, the reference image update determination circuit 25 detects pixels (zero pixels) whose magnitude of change is smaller than a set value R1 (32 for the threshold value in FIG. ) is smaller than a certain value R6, it is determined that there is a sign of illuminance change, and a command is given to the reference image updating circuit 24 to update the reference image to the current image. By the way, the histogram of the difference image between images with the illumination on, such as the photo in Figure 30, is the third one.
As shown in Figure 3, most of the pixels are zero pixels below the set value R1, and the number of zero pixels exceeds the set value R6.

Furthermore, in the case of intrusion detection examples such as the photographs in Figures 5 and 6, the histogram (Figure 7) of the current image (Figure 5) and the histogram (Figure 9) of the reference image (Figure 6) are There is almost no difference in shape or average value. This also applies when the auto iris (automatic gain adjustment circuit) of the TV camera 1 is working effectively. In the histogram of the difference image in this case (FIG. 11), the number of zero pixels below the set value R1 is greater than the set value R6, so it can be determined that there is no sign of illuminance change.

Note that when a monitoring area is set by the illuminance change monitoring area setting circuit 27 and the number of change detection grids in the area changes, it is necessary to normalize the frequency of the histogram and the set value R6 to the number of change detection grids. There is.

In this way, by detecting changes in illuminance using this histogram, false alarms due to changes in illuminance can be prevented without providing the illuminance meter 7.

Next, a method for determining signs of changes in illuminance without using the illuminance meter 7 or histogram will be described. This method uses a set value in the continuous change judgment circuit 10.
If it is determined that the number of continuous changes is greater than or equal to R7, it will be determined as an abnormal change and an alarm will be issued, and the set value
If the number of changes is less than R7, it is determined that there is a sign of illumination change, and a command to update the reference image is given to the reference image update circuit 24, thereby updating the reference image to the current image. This method of detecting signs of illuminance changes is effective when the entire image changes in a short period of time, such as when a light is turned on and off, but when the change is relatively gradual, such as a change in solar radiation, it is possible to The difference between the current image and the reference image may increase over a considerable period of time and may be judged as an abnormal change at some point. It is also necessary to calculate a moving average and use the moving average as a reference image, or to update the reference image at regular intervals.

The above methods for detecting signs of illuminance changes are also effective when the monitoring area is indoors and only a portion of the image is illuminated by sunlight entering through a window.
In particular, in the histogram method, the area of pixels used for illuminance change determination is set as the illuminance change monitoring area, and by setting the illuminance change monitoring area setting circuit 27 only in the position illuminated by sunlight, the difference in only this area can be set. The reference image may be normalized when it is determined that there is a change in illuminance based on the histogram of the image.

In addition, when there is a low possibility that only a part of the image is illuminated and the indoor lighting conditions are fixed, or when used indoors, the reference image or current image is normalized against changes in illuminance across the entire monitoring area. By doing so, it is possible to more accurately detect changes even in the case of gradual changes in illuminance.

Next, the change detection process of this embodiment will be explained. First, the pixel change test circuit 28 tests the magnitude of change for each pixel using a set value R1, and the projection calculation circuit 14 calculates the magnitude of only the pixels that have changed more than the set value R1 for each line parallel to the X and Y coordinate axes. The projection is calculated by adding up each time. Fig. 12 shows a binary image of changed pixels obtained by comparing the current image in Fig. 5 with the reference image in Fig. 6 and binarizing the changed pixels into 1 and the zero pixels into 0. Change detection grid (every 4 pixels)
The changed pixels are shown as black dots, and the outline of the intruder appears. Incidentally, FIGS. 8 and 10 show buffer images of the binary images shown in FIGS. 5 and 6. For binary images, the contour image is shaped by propagation and degeneration processing in known processing directions, and is used as an image feature for pattern recognition and alignment. It is also possible to obtain this, and it will be used in conjunction with the identification of intruding objects. Therefore, the illumination conditions that are important for identifying intruding objects are kept substantially constant by normalizing the illuminance, the reference value R1 for change pixels and zero pixels is clearly determined, and the binary pixel value of change pixels is This has the effect of making extraction easier, ensuring identification of intruding objects, fires, etc., and enabling image monitoring with fewer false alarms regarding abnormal change determination. Furthermore, since the correction coefficient for the background is applied to the intruding object, mismatching of the correction coefficients makes the pixel change of the intruding object noticeable, and it becomes easy to binarize the changed pixel. However, in order to verify an intruding object, correction is performed to match the intruding object. If there is a lot of noise and it is difficult to binarize the changed pixels, setting an upper limit on the size of the changed pixels will allow the size of the changed pixels to have a range and at the same time reduce the large difference due to isolated point noise. The influence of pixel value changes can be reduced.

FIG. 13a shows the projection on the X-axis of the cumulative value of the size of the changed pixel, and FIG. 13b shows the projection on the Y-axis. The dashed-dotted line in the figure indicates the setting value R2,
The two-dot chain line is the set value R5. This projection of the size of the changing pixel is based on the difference between the background and the intruding object, so it is not suitable for identifying the intruding object, but regarding the projection of the binary image of the changing pixel, when there is only one changing object, Helps identify objects. When there are a plurality of changing objects, the changing object is separated and cut out at the level of the binary image of changing pixels as shown in FIG. This will be useful for comparison and identification in the object comparison and identification circuit 16. Here, the changing object matching and identifying circuit 16 has a standard pattern 22 recorded in advance, and can perform comparing and identifying the changing object by comparing this standard pattern 22 with cutting data of the changing object and projection calculation data.
Furthermore, the values of the changed pixels in the difference image are replaced with the values of the corresponding pixels of the current image to calculate the projection, thereby making it possible to collate and identify the changed object in more detail. In other words, in the case of a fire, the value of the changed pixel increases (ignitability), so the strength of the projection also increases relative to the size of the changed area on the coordinate axes, and conversely, the strength of the projection of an intruder, etc. is small. Become. In other words, when calculating the projection on each coordinate axis, the value of the projection of the change pixel of the current image is divided by the cumulatively added number of change pixels (same as the projection of the binarized change pixel), and the change object per change pixel is calculated. If the brightness of the changing object per one changing pixel is abnormally high, it can be said that there is a high possibility of a fire. The brightness determination circuit 17 performs normalization and brightness determination for each pixel of this projection change.
It is.

The above explanation is for a monochrome image, but in the case of a color image, the images are compared for each color component of RGB, the changed pixels are binarized, and after the binarization, the binarized changed pixels are superimposed. Find changing pixels.

In addition, in order to strictly distinguish between a fire and an intruder, in this embodiment, the band filter 32 separates the infrared region and the visible light region, and simultaneously and separately A/D converts the light into the visible light region in the case of an intruder. The two are distinguished by the fact that the change is mainly in the light range, and in the case of flammable objects, the change in the infrared range is large, but incandescent light bulbs and sunlight also have a large change in the infrared range. By similarly normalizing illumination changes in the outside area, false alarms can be prevented.

Now, testing the change area by projection is less affected by noise such as isolated points of change pixels than determining the change range using a binary image of change pixels as shown in Figure 12, and changes can be clearly detected. It has the advantage of being possible. Furthermore, when there are a plurality of changed objects, the change detection area can be separated more accurately by cutting out the changed objects using a binary image of changed pixels and then calculating the projection.

Here, in the change area projection verification circuit 18, the range on the coordinate axes of the projection beyond the set value R2 is determined for each coordinate axis in the change detection region verification circuit 19 ((X ₁₁ , X ₁₂ )...(Xn ₁ , Xn ₂ ) , (Y ₁₁ , Y ₁₂ )…
(Yn ₁ , Yn ₂ ), (Xi ₁ , Xi ₂ )(Yi ₁ ,
The size of the rectangular region determined by Yi ₂ ) is the set value
It is determined that there is a change when it is larger than RX3, RY3,
Settings for projection by change area calculation circuit 20
The section exceeding R5 is determined as a changing region, and the image of this rectangular region is appropriately compressed and encoded by the image information compression/transmission circuit 21 and transmitted to the receiver side through a telephone line or the like. Also, change area projection verification circuit 1
The test data of 8 is the change object movement determination tracking circuit 23.
is input together with the changed object identification data from the changed object matching and identification circuit 16, and the changed object is moved and tracked by the changed object movement determination and tracking circuit 23.
Among the bright changing objects determined by the brightness/darkness determination circuit 17, those that are stationary are determined to be fire ignition locations, relatively dark changing objects, and moving objects are determined to be intruders. Here, the projection size of the changing pixel of the current image in the infrared region obtained by photographing through the band filter 32 in the infrared region is verified, and the size is determined by If the size is larger than the projected size of the changing pixel in the visible light region obtained by photographing, it is determined to be a heat-generating changing object, and in the opposite case, it is determined to be a heat-receiving changing object. becomes certain.

By the way, if you want to send only the changed pixels instead of the rectangular area, set value R4 (R4 < R1) for changing pixel judgment.
Then, changed pixels are determined again and changed pixel information of the current image is transmitted.

The above-mentioned detection of signs of illuminance changes, selection and combination of normalization methods, and parameter settings are optimally set according to the environmental conditions of the monitoring site and the monitoring target. These settings are made by the illuminance change detection method setting circuit 6.
Alternatively, the illuminance change monitoring area setting circuit 27 performs this.

17 and 19 show buffer images of the binary images of the current image and the reference image in FIGS. 14 and 15, and FIG. 21 shows a buffer image of the difference image. Moreover, FIGS. 22a and 22b show the X and Y projections of the difference image. Furthermore, Figure 23~
Figures 29a and 29b are the histogram of the current image in Figure 5, the buffer image of its binary image, the histogram of the reference image in Figure 6, and the buffer image of its binary image when the monitoring area is defined, respectively. , a buffer image of the difference image between these images, and the X and Y projections of the difference image.

Also, Fig. 31 shows the 30th image with no change in illumination and no intrusion.
The histogram of the image in the figure, Figure 32 is the buffer image of the binary image, Figure 34 is the buffer image of the binary image of the difference image between Figures 30 and 35.
Figures a and b show the buffer image of the difference image and the X and Y projections of the difference image.

Note that smoothing may be applied to the projection. This calculates the projection of each point as the average value of the previous and subsequent projections,
This is done by weighting the values of the previous and subsequent projections and then finding the average value. This smoothing can remove small changes.

FIG. 3 shows the main circuit configuration of an embodiment that is a modification of the embodiment shown in FIG. The calculation circuit 26 calculates the magnitude of change for each pixel between this normalized image and the current image to obtain a difference image.The difference image is stored in the difference image frame buffer 30, and the pixel changes are stored in the difference image frame buffer 30. The test circuit 28 replaces the value of the changed pixel with the value of the corresponding pixel of the current image, stores it in the changed pixel frame buffer 31, and stores it in the changed pixel frame buffer 31 and the difference image frame buffer 30 and the changed pixel frame buffer 3.
Based on the stored data of 1, the projection calculation circuit 14 can selectively calculate projection of a difference image, projection of a binary image of the difference image, or projection of changed pixels.

[Effects of the Invention] The present invention verifies the magnitude of change for each pixel of the difference image between the current image and the reference image using a first set value, and determines the magnitude of change of the changed pixel that has changed by more than the set value. In an image monitoring method that calculates the projection by cumulatively adding up each line parallel to the X and Y coordinate axes, the infrared region and visible light region are separated by a bandpass filter and the image signals of each region are calculated. A/D conversion is performed to determine the projection of each changing pixel in the above process, and when the projection of the changing pixel in the infrared region is larger than that in the visible light region, it is considered an exothermic changing object; Since there is a process in which it is determined as a heat-receiving changing object when the projection of a changing pixel in the light area is large, it is possible to reliably distinguish between changing objects as fire and non-fire, making it easier to identify fires and intruders, thereby preventing false alarms. Furthermore, since an image in the infrared region is used, it is easy to confirm the ignition location in the smoke.

[Brief explanation of the drawing]

FIG. 1 is a circuit configuration diagram of an embodiment of the present invention, FIG. 2 is an illuminance normalization calculation circuit same as above, FIG. 3 is a main circuit diagram of another embodiment of the change detection process, and FIG. 4 is a TV Examples of camera arrangement, Figures 5 to 35 are explanatory diagrams of the same operation as above, 1 is a TV camera, 14
15 is a projection calculation circuit, 15 is a changing object matching identification circuit,
17 is a brightness determination circuit, 18 is a change area projection calculation circuit, 19 is a change detection area verification circuit, 20 is a change area calculation circuit, and 32 is a band filter.

Claims (1)

[Claims] 1. The magnitude of change in each pixel of the difference image between the current image and the reference image is tested using a first set value, and the magnitude of change in the changed pixel that has changed by more than the set value is determined. X, Y
In the image monitoring method, which has a process of cumulatively adding up each line parallel to the coordinate axes of In the above process, the projection of each changing pixel is calculated, and when the projection of the changing pixel in the infrared region is larger than that in the visible light region, it is considered an exothermic changing object. An image monitoring method characterized by having a process of determining a heat-receiving variable object when the projection of a pixel is large. 2. The image monitoring system according to claim 1, characterized in that images in the infrared region and visible light region are obtained using the same TV camera by switching band filters.