JPH07262355A - Image monitoring device - Google Patents

Abstract

PURPOSE:To stably extract a monitoring object body even when the background is changed by the change of sunshine or the movement of clouds. CONSTITUTION:This device is provided with a block B1 for detecting the kinds of changes in the background by comparing a dictionary subjected to the learning processing of the projection feature amount of estimated changes in the background with the projection feature amount of the actual change, block B2 for deciding the still degree of the monitoring object body, background change integrally deciding means 23 and background updating means 26 or the like. When the still degree of the monitoring object body is higher than a prescribed value, the background is updated while including the monitoring object body but when the overlap degree of a large area except the monitoring object body is larger than a prescribed value, the background is updated while including that large area. Thus, even when the background is changed, stable monitoring is enabled.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image monitoring apparatus for monitoring a dangerous area or the like from an image picked up by an image pick-up means such as a television camera, detecting an object to be watched and outputting an alarm.

[0002]

2. Description of the Related Art Conventionally, in an image monitoring apparatus which uses an image obtained by an image pickup means such as a television camera, a difference between a background image and a monitoring image previously created for detecting an intruding object is calculated, A method of detecting an intruding object or an abnormal state in a monitor image by utilizing the feature amount of the changed portion of is known.

In the conventional method as described above, an intruding object or an abnormal state can be accurately detected when there is no change in the background image, but an object having a temporal change in the background (cloud movement, tree shake, etc.). ), Or if there is a change in the background itself (change in sunshine, lighting in the room, turning off, etc.), fluctuations other than the target to be detected are extracted in the difference image between the background image and the monitoring image. , The monitoring target may not be extracted properly. As a result, there is a drawback in that the reliability of the system is reduced due to a detection error or a false detection of the monitored object.

Therefore, in order to eliminate this drawback, for example, the following methods (1) and (2) have been proposed. (1) When it is detected that the background image has changed so that the target object can be reliably detected from the difference image, the background image is updated at an appropriate timing. As an example of a method of detecting a change in the background image, the duration of the value is counted for each pixel, and when the duration exceeds a certain threshold,
The background of the pixel is updated (Japanese Patent Laid-Open No. 4-11107).
No. 9).

(2) As a background image, an image including steady changes of (a) average and variance (b) maximum value and minimum value of background images of several screens is created, and the input image is created. Is output as a difference image only when it takes a value outside the allowable range determined by these (JP-A-4-330576).
(See the official gazette).

[0006]

However, the above method has the following problems. Above (1)
In the method, the background image is not updated until the change of each pixel with respect to the initial background image continues for a certain period or more when the background image is updated, because the image change is monitored in pixel units. Therefore, there is a problem that even in the same object in the background, pixels that are updated and pixels that are not updated are mixed due to the fluctuation of the video level or the like.

Further, in the above method (2), since the average, variance, maximum or minimum is obtained for each pixel, it is effective for steady and periodic background changes, but it is effective for illuminance changes and illumination of the entire background image. If you create a background that includes changes in conditions, the variance will be large, or the maximum and minimum values will be too far apart and the allowable range will be expanded, making it difficult to detect the monitored object. Need control.

There are various situations in which the background needs to be updated, such as changes that require background updating, steady changes that do not necessarily require updating, and changes due to other objects. Not aware of changes, it is difficult to adapt to different contexts of background changes. in addition,
If the object to be monitored exists in the screen, the above (1),
Both methods (2) have a problem that it is difficult to update the background at an accurate timing. Therefore, an object of the present invention is to detect the type of background change by utilizing the knowledge about the change of the background image and update the background image only when necessary so that the monitoring target object can be stably extracted. It is in.

[0009]

In order to solve such a problem, according to the invention of claim 1, an image pickup means for picking up an image of the monitoring area, a monitoring image memory for storing the imaged image data,
A background image memory that stores a background image, a difference / binarization unit that binarizes the difference between a time-series monitoring image and the background image, and a difference / binarization that stores the difference / binarized image. An image monitoring apparatus, comprising an image memory, for extracting a change area from the difference / binarized image and detecting a monitoring target object from the size, shape, and movement of the change area, First extraction means for obtaining each differential projection value in the horizontal direction and extracting the feature amount, and differential projection of the area and degree of change of the background image with respect to some expected variations of the background image. A dictionary memory that learns as a value feature quantity and stores the result as a background projection feature quantity dictionary and a projection feature quantity obtained from the difference / binarized image of the monitoring image are compared with a background projection feature quantity dictionary, and Belongs to the expected type of background change, Determination means for determining other than the above, second extraction means for extracting the monitoring target object from the difference / binarized image of the monitoring image, and background updating in accordance with the background change within a fixed time and the object extraction result. It is characterized in that it is provided with an updating means for carrying out, and stable monitoring is possible even if the background changes.

According to the first aspect of the invention, associating means for associating a temporal object position with an area extracted as the object to be monitored, and the object position is stationary within an object extraction error range. A stillness degree calculating means for calculating the degree of staticity as a stillness degree, and for all the monitoring target objects extracted in the image, if the stillness degree is a certain value or more, the background is updated to include the stationary monitoring target object. When the object to be monitored is stationary, it is possible to perform stable monitoring including the area when the object to be monitored is stationary (claim 2).

According to the second aspect of the invention, with respect to the change area other than the area extracted as the monitoring target object, the area integrating means for integrating the individual change areas other than the monitoring target object into a larger adjacent area. And a degree-of-overlap calculation means for calculating the temporal degree of overlap of the area, and if the degree of overlap for all areas is a certain value or more, it is determined that the area is stationary, and the background is updated including the stationary area. A third updating means is added, and for a change that cannot be determined as a background change or a target monitored object, the background is updated when a stationary state continues for a certain period of time as a large area, and stable monitoring is enabled. It is possible (claim 3).

According to the second aspect of the present invention, the extracted area of the object to be monitored or an area creating means for expanding the area by a certain size to create a mask area, and a temporal object of the object to be monitored. A judgment means for performing a logical product operation of a plurality of mask areas according to the movement to judge whether or not the mask areas remain, and a background image creation for creating a background image in which the mask areas are filled with each other from a plurality of background images with masks. By adding a means, the background can be updated even when the monitored object moves in time, and stable monitoring can be achieved (claim 4).

[0013]

[Operation] First, some background fluctuations expected at a certain position (background changes that require background updates such as sunshine changes, door opening and closing, and steady background changes such as cloud movements and tree shakings) For the degree of change and the changing area that appear on the image,
The projection values in the vertical (X-axis) direction and the horizontal (Y-axis) direction of the difference / binarized image are obtained, and these are learned in advance as projection feature quantities to create a background projection feature quantity dictionary.

Thereafter, the projection feature amount is extracted from the difference / binarized image between the monitor image and the background image during actual monitoring,
By comparing this with the background projection feature quantity dictionary, it is determined what kind of background image change the image change is, and other (unknown), and the background update processing is performed according to the result. Even if the monitoring target object is extracted, if it is determined that the object is stationary for a certain time or more within the range of the object extraction error, the extracted monitoring target object is updated as the background to detect the vicinity of the stationary monitoring target object. Be able to monitor the changes that appear in.

In addition, with respect to an expected background change and an unknown image change other than the object to be monitored, if the change is considered to be a large unit of a certain degree or more and is considered to be stationary, the change will occur. Update changes as background changes. For example, parking a car or leaving a large baggage
This processing enables detection. Furthermore, even when there is a moving target object in the monitoring image, an image in which the part of the monitoring target object is masked is created, and the process of filling the masked areas with each other with multiple masked images is performed. By creating and updating images, it is possible to quickly deal with background fluctuations.

[0016]

1 is a block diagram showing an embodiment of the present invention. In FIG. 1, reference numeral 1 is an image pickup means such as a television camera (here, an A / D converter is also included), 2 is a monitor image memory of a monitor image input by the image pickup means 1, 3
Is a background image memory storing background images, 4 is a difference / binarization means for performing difference / binarization between the monitor image and the background image,
Reference numeral 5 denotes a differential / binarized image memory for storing the result.

Reference numeral B1 denotes a block for determining the type of background change, which is composed of 6-12. 6 is difference / 2
Projecting means for projecting the binarized image in the vertical and horizontal directions,
Reference numerals 7 and 8 denote projection value memories that store projection values in the vertical direction (X axis) and horizontal direction (Y axis), 9 is extraction means for extracting projection feature amounts from the projection value memories 7 and 8, and 10 is an extraction unit. It is a projection feature amount memory that stores the projection feature amount extracted by the means 9. Reference numeral 11 denotes a determination unit that compares the projected feature amount with a background projected feature amount dictionary 12 created by learning with respect to an expected background variation and determines the type of background change.

Reference numeral B2 denotes an object stillness determination block for determining the stillness of the object to be monitored and other large areas, and it is composed of 13 to 21. Reference numeral 13 is an object extracting means for extracting the monitoring target object from the differential / binarized image memory 5, 14 is an object extraction result extracted by the object extracting means 13, 16 is an object extraction result at time t = T, and The associating unit 17 associates the object extraction result 15 in the frame (t = T-1) with time, and 17 is a stillness determining unit that calculates and determines the stillness of the target object from the association result.

Reference numeral 18 denotes an unifying means for unifying the change areas other than the target object for each large area nearby, and the data of the change areas integrated over a certain area are stored in the memory 19. Reference numeral 21 is a degree-of-overlap determination means for determining and determining the degree of overlap between the changed area data 20 integrated in the previous frame and the changed area data 19 in the current frame.

Reference numeral 22 is a mask image creating means for creating a background image in which only the part of the monitored object is masked when the monitored object appears in the monitored image memory 2, and 23 is a background change type determination result, a target object extraction result. The background change comprehensive determination means for comprehensively determining the degree of stillness of the target object and the degree of change area overlap and determining the background update mode are shown.

Numeral 24 is an abnormality judging means for comprehensively judging the object extraction result, background change, etc. to detect the appearance or abnormal state of the monitored object and outputting an alarm, and 25 is a background image buffer. The background image buffer 25 may have as many frames as the number of frames from when the background change is detected until the actual update, or when the memory is small, the background image buffer 2
5 and the calculation result such as the average value between the monitoring image memory 2 may be stored in the background image buffer 25.
Further, 26 is a background updating means for creating a background image from the background image buffer 25.

FIG. 2 shows an example of the differential / binarized image. Figure 2
(A) shows an example of a differential / binarized image and its projection pattern when the brightness of the entire background image changes due to changes in sunshine, changes in illumination, and the like. In such a case, since the brightness fluctuates on the entire screen, a pattern appears on the entire differential / binarized image. Further, FIG. 2B is an example of the differential / binarized image in the case where there is no change in the background and there is a monitoring target object. In this case, the pattern of the monitored object appears in the differential / binarized image. Normally, since the monitored object does not appear uniformly on the entire screen, the projection values of the differential / binarized image in this case appear in a cluster on the monitored object.

Here, an example of expected background change will be shown. Background change C: Change of sunshine (outdoor), change of lighting (indoor) Background change D: Shake of trees and movement of clouds Background change E: Stop of car (in case of fixed position) Background change F: Object moved by person ( Doors, windows, etc.)

Among the above, in the case of C, a pattern appears in the entire difference / binarized image, and monitoring cannot be continued as it is. Therefore, it is necessary to quickly update the entire background. However, in the case of D, it is a periodic background fluctuation that appears partially and does not need to be updated. In the cases of E and F, the background is partially irregular, and in this case, the background may be partially updated only when the change continues for a certain time.

FIG. 3 is an explanatory diagram for explaining an example of the differential projection value and its feature amount, and here, only the X-direction component is shown. Here, the differential projection value HX as shown in FIG.
(I) (i = 1, 2, ..., ISX) and HY (j)
Assuming that (j = 1, 2, ..., ISY) is obtained, the projection feature amount extraction means 9 extracts the following projection feature amounts. And THX represents a threshold value (see FIG. 3).

(A2) X-direction difference value appearance rate RX = SUMX / I
SX 100 (%) (a3) Difference value appearance area X lower limit value XMIN (a4) Difference value appearance area X upper limit value XMAX (a5) Difference value appearance area number XN

(A6) X difference value average HXAV HXAV = Σ HX (i) / SUMX HHX (i) ≠ 0 (a7) X difference value variance HXSIG HXSIG = Σ (HX (i) −HXAV) ² / SUMX HHX ( i) ≠ 0

Data similar to the above is extracted also in the Y direction. And Note that THY indicates a threshold value.

(B2) Y-direction difference value appearance rate RY = SUMY / I
SY · 100 (%) (b3) Difference value appearance area X lower limit value YMIN (b4) Difference value appearance area X upper limit value YMAX (b5) Difference value appearance area number YN

(B6) Y difference value average HYAV HYAV = ΣHY (j) / SUMY HHY (j) ≠ 0 (b7) Y difference value variance HYSIG HYSIG = Σ (HY (j) −HYAV) ² / SUMY HHY ( j) ≠ 0

FIG. 4 is a flow chart for explaining the background change type determination operation. I-1) When no pattern appears in the differential / binarized image at time t = T, the background update mode A is set (no background change, no target object). When a pattern appears in the difference / binarized image, the projection values HX (i) (i = 1, 2, ..., ISX), HY of the difference / binarized image
From (j) (j = 1, 2, ..., ISY), the above (a1
~ A7) and (b1 to b7), a total of 14 projection feature quantities are obtained.

Here, in the case where only a pattern having a small amount of noise appears in the differential / binarized image, the background update mode B is set as a case similar to the case of I-1). As a method of determining noise, for example, the following four conditions of the projection feature amount may be used. SUMX <SUMXMIN and RX <RXMIN and SUMY <SUMYMIN
And RY <RYMIN

II) When a pattern other than noise appears in the differential / binarized image, the calculated projected feature amount is compared with the contents of the background feature amount dictionary 12, and any one of the background changes C to F, or If it is judged that it is not close to any background change, it is unknown. As a comparison method with the dictionary 12 in this case, the difference between the respective feature amounts may be simply normalized and then added to obtain a small addition value, or a vector inner product (cos θ) in the conventional feature space may be used. . However, for all features, only those that are similar are regarded as specific background fluctuations, and others are unknown.

FIG. 5 is a flow chart showing a method of determining the stillness of an object. III) Even when compared with the projected feature quantity of the predicted background variation,
When it is determined that there is no background fluctuation that matches and it is determined that the object is unknown, the monitoring target object is extracted. The extraction of the object to be monitored here is performed in the same manner as the conventional method. The stationary degree of the monitoring target object is stored in the flag OBJ, and the overlapping degree of the area other than the monitoring target object is stored in the flag AREA. In both flags, "0" indicates no object, "1" indicates stationary, and "2" indicates moving.

IV-1) When the object to be monitored is extracted, the position of the object is calculated, and the position of the object is correlated with the position of the object in the previous frame to calculate the stillness St (t) of the object. An example of a function representing the stillness of an object is shown by the following expression (1). However, Ustn is a range of object extraction error, Ut is a distance between target objects associated in the past, and MAXFRM is a constant indicating time. When the associated time is MAXFRM, St (t) becomes “1” when the position of the associated object does not move at all, and becomes smaller as the moving range becomes wider. Further, even if the associated object is stationary from the middle, it takes a larger value as the time after stationary becomes longer.

FIG. 6 is an explanatory diagram for concretely explaining the stillness determination of the object to be monitored. FIG. 6A shows the object position association result (t = T-3 to T). Here, it is assumed that the object 1 is moving and the object 2 is almost stationary. (B) is an example of a monitoring target object table, (c) is a graph showing the relationship between the associated time and stillness. In this graph, the static degree St is initially set for both the objects 1 and 2.
Although (t) is 0, the moving object 1 does not have a large St (t) even when the associated time elapses, whereas the stationary object 2 has a monotonic St (t) with time. It has been shown to increase.

Therefore, for all the extracted objects to be monitored, when St (t) ≧ SMIN (threshold value) is satisfied, the object to be monitored is considered to be stationary, and the flag OBJ = 1 is set. Otherwise, the flag OBJ is set. = 2.

IV-2) Next, the change areas excluding the object to be monitored are grouped into large change areas. As an example of a method of grouping, a method may be considered in which the difference / binarized image is expanded by several pixels and then expanded, and nearby patterns are integrated into one. If there is no area having a certain area or more as a result of integration, AREA = 0.

When a region having a certain area or more is formed by the integration, the position and size of this region are calculated, and the overlapping degree of the regions is determined using the function Ot (t) to determine whether or not the region is still. An example of the function Ot (t) is shown in the following expression (2).

A (t) in the above equation (2) is the total area of the regions integrated at time t, and AND (t-1, t) is the region at time t-1 and the region at time t. The area of the overlapping region is shown. FIG. 7 is an explanatory diagram for explaining a specific example of the overlapping degree determination. In the figure, (a) shows the areas 1, 2, 3 and their areas A1, A2, A3 at time t-1, and (b) shows the areas 1, 2, 3 and their areas A1 ', A2 at time t. ',
A3 'and (c) indicate AND (t-1, t), respectively.

When the areas integrated at the time t-1 and the time t are completely overlapped as in the area 2 in FIG. 7, the degree of overlap Ot (t) = 1 obtained by the above equation (2) is Area 3
When there is no overlap as in, the degree of overlap Ot (t)
= 0. That is, the degree of overlap monotonously decreases from the state of completely overlapping to the state of not completely overlapping, depending on the degree of overlapping. Therefore, if Ot (t) ≧ OMIN (threshold value), the area is considered to be stationary and AREA =
1 and AREA = 2 otherwise. α is t-1,
It is a constant indicating the weight of information between t.

FIG. 8 is a flow chart showing the processing of the background update comprehensive judgment section. (1) Judgment of the background update mode, background changes D, H, K,
If it is determined to be N, the background update is not necessary, or is determined to be impossible, and the background update is not performed. (2) Background changes A, B, C, E, F, G, I, J, L,
In the case of M, the following processing is performed.

(2-1) When it is first determined that the background change is of that type, the background change duration f = 1 is set, and (2-1-1) In the case of the background changes L and M, the monitored object is selected. The number of pixels in the appropriately expanded mask area is 1, and the others are 0.
The mask area image is copied to the buffer MASK, and the background image in which the monitoring target object area is masked from the monitoring image is copied to the background image buffer FRAME (f). The background updatable flag FLG is initially OFF.
Set to. (2-1-2) In other cases, the contents of the monitor image memory 2 are copied to the background image buffer FRAME (f).

(2-2) When the same type of background change continues for two or more frames, (2-2-1) f = f + 1. (2-2-2) If the background changes L and M, copy the masked background image to the background image buffer FRAME (f),
The logical product of the mask area image and the mask image buffer MASK is calculated, the result is stored in the mask image buffer MASK, and it is checked whether or not the pixel having the value 1 remains in the MASK. The flag FLG is turned ON at the time when there is no pixel with a value of 1 in MASK. (2-2-3) In other cases, the contents of the monitor image memory 2 are copied to the background image buffer FRAME (f). (2-2-4) When f has reached the number of frames required for background update and the background changes L and M, the flag FLG is ON.
Only when is set, the background is updated.

(3) As a background updating method, a background is created from the background buffer 25 within a fixed time (the same as the conventional method). However, for the masked background images of the objects to be monitored, the masked portions are filled up with each other to create the background. The relationship between the background update mode and the background update processing in each case is summarized as shown in FIG.

FIG. 10 is an explanatory diagram for explaining the masking image padding process. Now, assuming that the two background image buffers are IMG1 and IMG2 and the mask pixel value is M, IMG (X, Y) = {IMG1 (X, Y) + IMG2 (X, Y)} / 2 (however, IMG1 ( X, Y), IMG2 (X, Y) ≠ M, or IMG1 (X, Y), IMG2 (X, Y) = M) = IMG1 (X, Y) (where IMG1 (X, Y) ≠ M, IMG2 (X, Y) = M) = IMG2 (X, Y) (However, when IMG1 (X, Y) = M, IMG2 (X, Y) ≠ M) (3) By performing the above, it is possible to create an average image between two masked background images. If this process is performed until the pixels of the value 1 are eliminated in the mask image buffer MASK, it becomes possible to create a background image that is not affected by the monitored object.

[0047]

According to the present invention, when the change area is extracted from the difference between the image input through the image pickup means and the background image, and the monitoring target object is extracted from the change area, the expected change in the background is expected. By comparing the dictionary that learned the projection feature amount of and the projection feature amount of the actual change, the type of background change is detected, and (b) the stillness of the monitored object is calculated. Update the background. (B) Obtain the degree of overlap of large areas other than the object to be monitored,
If it is larger than the predetermined value, the background is updated. (C) When the monitored object moves, a background image is created by masking that part, and the background is updated when the masked area disappears. Since the above process is performed, there is an advantage that stable monitoring can be performed even when the background changes.

[Brief description of drawings]

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

FIG. 2 is an explanatory diagram for explaining a relationship between a difference / binarized image and a projection value.

FIG. 3 is an explanatory diagram showing an example of a differential projection feature amount.

FIG. 4 is a flowchart showing a background change type determination operation.

FIG. 5 is a flowchart showing an object stillness determination operation.

FIG. 6 is an explanatory diagram for describing a specific example of object stillness determination.

FIG. 7 is an explanatory diagram for describing a specific example of area overlapping degree determination.

FIG. 8 is a flowchart showing a comprehensive determination operation.

FIG. 9 is an explanatory diagram illustrating a relationship between a type of background variation and update processing.

FIG. 10 is an explanatory diagram of a padding process between masked background images.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Imaging means, 2 ... Monitoring image memory, 3 ... Background image memory, 4 ... Difference / binarization means, 5 ... Difference / binarization image memory, 6 ... Projection means, 7 ... X projection value memory, 8 ... Y projection value memory, 9 ... Projection feature extraction means, 10 ... Projection feature memory, 11 ... Background change type determination means, 12 ... Background projection feature amount dictionary, 13 ... Object extraction means, 16 ... Target object position associating means, 17 ... Target object stillness determination means, 18 ... Change area integrating means, 19 ... Integrated change area data, 20
... Integrated change area data in previous frame, 21
... change area overlap degree judging means, 22 ... mask image creating means, 23 ... background change comprehensive judging means, 24 ... intrusion detection / abnormality judging means, 25 ... background image buffer, 26 ... background updating means, B1 ... background change type judgment Block, B2 ... Object stationary determination block.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification number Reference number within the agency FI Technical indication H04N 7/18 D (72) Inventor Akira Shimizu 1 Fujimachi, Hino-shi, Tokyo Fujifacom Systems Co., Ltd. Within

Claims (4)

[Claims] 1. An image pickup means for picking up a surveillance area, a surveillance image memory for storing imaged image data thereof, a background image memory for storing a background image, and a difference between a time-series surveillance image and the background image. The difference / binarization means for binarizing and the difference / binarized image memory for storing the difference / binarized image are provided, and the change area is extracted from the difference / binarized image, and the size of the change area is set. An image monitoring apparatus for detecting an object to be monitored from a shape and a motion, first extracting means for obtaining respective differential projection values in the vertical and horizontal directions of the differential / binarized image, and extracting a feature amount thereof, A dictionary memory that learns the area and degree of change of the background image as a difference projection value feature quantity in response to some expected changes in the background image, and stores the result as a background projection feature quantity dictionary; Obtained from image difference / binarized image And comparing the projected feature quantity with the background projected feature quantity dictionary to determine whether it belongs to an expected type of background change or not, and monitoring from the difference / binarized image of the monitoring image. A second extracting means for extracting the target object and an updating means for updating the background in accordance with the background change and the object extraction result within a fixed time are provided to enable stable monitoring even if the background changes. An image monitoring device characterized by. 2. An associating unit for associating a temporal object position with an area extracted as the object to be monitored, and a degree that the object position is stationary within an object extraction error range is static. And a second updating unit for updating the background including the stationary monitoring target object when the stationary degree is a certain value or more for all the monitoring target objects extracted in the image. The image monitoring apparatus according to claim 1, wherein when the object to be monitored is stationary, stable monitoring is possible including the area of the object. 3. An area integration unit for integrating individual changed areas other than the object to be monitored into a larger area adjacent to the changed area other than the area extracted as the object to be monitored, and temporal overlap of the areas. And a third updating means for updating the background including the still area when it is determined that the area is stationary when the overlapping degree is equal to or more than a certain value for all areas. In addition, for a change that cannot be determined as a background change or a target monitoring object, the background is updated when the stationary state continues for a certain period of time as a large area, and stable monitoring is possible. The image monitoring device according to 2. 4. A region creating means for creating a mask region by expanding the extracted region of the monitoring target object or the region by a certain size, and a logic of a plurality of mask regions according to temporal movement of the monitoring target object. Addition of judgment means for performing a product operation to judge whether or not a mask area remains, and background image creating means for creating a background image in which mask areas are filled with each other from a plurality of background images with masks The image monitoring apparatus according to claim 2, wherein the background is updated even when the object is temporally moving to enable stable monitoring.