JPH11296654A - Image monitoring device and image monitoring method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect change with excellent accuracy at all times without being affected by the change of weather and seasons relating to an image monitoring device for detecting the change inside a monitoring area monitored by an image sensor. SOLUTION: The image sensor 10 outputs image data constituted of plural pixel values and an area division part 12 appropriately divides the image data and generates area data. A change evaluation value calculation part 24 computes a change evaluation value for indicating the certainty of assuming that the distribution of the pixel values included in the area data is constituted of the set of normal values and the set of change values. A non-change evaluation value calculation part 26 computes a non-change evaluation value for indicating the certainty of assuming that the distribution of the pixel values is constituted of only the set of the normal values. A change judgement part 48 compares the change evaluation value and the non-change evaluation value and judges the presence/absence of the change.

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 and an image monitoring method, and more particularly to an image monitoring apparatus and an image monitoring method suitable as an apparatus and method for detecting a change in a monitoring area monitored by an image sensor. About.

[0002]

2. Description of the Related Art Conventionally, Japanese Patent Application Laid-Open No.
As disclosed in Japanese Unexamined Patent Publication No. 6-206, there is known an apparatus that monitors a predetermined area by an image sensor and detects occurrence of an abnormality in the area. The above-described conventional apparatus extracts a feature of a temporal or spatial change from image data acquired by an image sensor, and when it is determined that a significant change is recognized in the image data based on the feature, an abnormality in the monitoring area is determined. Is detected. According to the above method, it is possible to easily detect an abnormality that occurs in the monitoring area.

[0003]

The image data can be understood as a set of a plurality of pixel values arranged two-dimensionally. Therefore, whether or not the image data has changed can be determined, for example, by determining whether or not some of the plurality of pixel values have changed. Whether or not a change has occurred in some pixel values is determined by, for example, determining whether a change from a normal value has occurred for each pixel value.
Alternatively, it can be determined by determining whether the distribution of the plurality of pixel values has changed from the normal distribution.

Therefore, in the above-described conventional apparatus, it is determined whether or not each of a plurality of pixel values constituting the image data has changed from a normal value, or the distribution of the plurality of pixel values is determined. By determining whether or not there is a change from the normal distribution, it is possible to determine whether or not there is a change in the image data, that is, whether or not there is an abnormality in the monitoring area.

[0005] However, the value of the pixel value constituting the image data is greatly affected by the state of the air interposed between the monitored object and the image sensor, the intensity of light applied to the object, and the like. . For this reason, the normal value of each pixel value and the normal distribution of a plurality of pixel values show changes according to the weather, the season, and the like even when the state of the target object is constant. Therefore,
It is difficult to always detect a change in an object with excellent accuracy by a conventional method that requires a normal value or a normal distribution to be determined in advance.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problems, and is an image monitoring apparatus that always detects a change in a monitoring area with excellent accuracy without being affected by changes in weather or season. The first object is to provide

It is a second object of the present invention to provide an image monitoring method for always detecting a change in a monitoring area with excellent accuracy without being affected by changes in weather or season.

[0008]

According to a first aspect of the present invention, there is provided an image monitoring apparatus, comprising: an image sensor capable of outputting a plurality of pixel values constituting image data of a monitoring area; and a distribution of the plurality of pixel values. A distribution detecting means for detecting, and a change evaluation value representing a certainty of assuming that the distribution of the pixel values is assumed to be a mixture distribution composed of a set of normal values and a set of change values changed from the normal value. Change evaluation value calculating means, and a non-change evaluation value calculating means for calculating a non-change evaluation value representing a probability of assuming that the distribution of the pixel values is a non-change distribution composed of only the set of normal values. And determining means for determining whether or not a change from a normal state has occurred in the monitoring area based on the change evaluation value and the non-change evaluation value.

According to a second aspect of the present invention, there is provided an image monitoring apparatus comprising:
Reference image data storage means for storing reference image data corresponding to a reference state of the monitoring area; difference image creation for creating a difference image corresponding to a difference between the image data acquired by the image sensor and the reference image data And means,
The distribution detecting means detects a distribution of a plurality of pixel values constituting the difference image.

According to a third aspect of the present invention, there is provided an image monitoring apparatus comprising:
The change evaluation value calculation means, each of the plurality of pixel values, a normal value set assuming the set of the normal values,
Classification means for classifying the set of change values and a set of assumed change values, and when it is determined that a change has occurred in the monitoring area, a set of parts corresponding to pixel values belonging to the set of change values is determined. A change range specifying means for specifying a change range is provided.

[0011] An image monitoring apparatus according to claim 4 of the present invention comprises:
Classification changing means for appropriately changing the classification by the classification means, a distribution of pixel values belonging to the set of normal values and a distribution of pixel values belonging to the set of change values, and a degree of conformity with a predetermined distribution. Classification evaluation means for evaluating the suitability of classification by the classification means, wherein the evaluation value calculation means calculates the change evaluation value based on a mixture distribution having a classification obtained by the classification evaluation means with an optimum evaluation. Is what you do.

According to a fifth aspect of the present invention, there is provided an image monitoring apparatus comprising:
The type of distribution of the plurality of pixel values includes a distribution type identification unit that identifies one of a plurality of distribution types assumed in advance, and the change evaluation value calculation unit and the non-change value calculation unit include the distribution type identification unit, The distribution type specified by the distribution specifying means is recognized as the type of distribution constituted by the set of normal values, and the change evaluation value or the non-change evaluation value is calculated.

According to a sixth aspect of the present invention, there is provided an image monitoring apparatus comprising:
The change evaluation value calculation means and the non-change value calculation means, respectively, is a method of calculating a predetermined reference value that enables superiority comparison between models having different numbers of free parameters, based on the plurality of pixel values, The change evaluation value or the non-change evaluation value is calculated.

An image monitoring method according to a seventh aspect of the present invention comprises:
From the image sensor, a pixel value detecting step of obtaining a plurality of pixel values constituting image data of the monitoring area, a distribution detecting step of detecting a distribution of the plurality of pixel values, and a distribution of the pixel values, A set, a change evaluation value calculation step of calculating a change evaluation value representing a probability of assuming a mixture distribution composed of a set of change values changed from the normal value, and a distribution of the pixel values,
A non-change evaluation value calculation step of calculating a non-change evaluation value representing the probability of assuming a non-change distribution composed only of the set of normal values, and the change evaluation value and the non-change evaluation value And determining whether the monitoring area has changed from a normal state based on the monitoring area.

An image monitoring method according to claim 8 of the present invention comprises:
A reference image data storage step of storing reference image data corresponding to a reference state of the monitoring area; and a difference image creation for creating a difference image corresponding to a difference between the image data acquired by the image sensor and the reference image data. And detecting the distribution of a plurality of pixel values constituting the difference image.

According to a ninth aspect of the present invention, there is provided an image monitoring method comprising:
The change evaluation value calculating step includes a classification step of classifying each of the plurality of pixel values into a normal value set assuming the set of normal values, and a change value set assuming the set of change values. When it is determined that a change has occurred in the monitoring area, a change range specifying step of specifying a set of parts corresponding to pixel values belonging to the set of change values as a change range is provided.

According to a tenth aspect of the present invention, in the image monitoring method, a classification changing step of appropriately changing the classification in the classification step, and a distribution of pixel values belonging to the set of normal values and belonging to the set of changed values. A distribution of pixel values, and a classification evaluation step of evaluating the suitability of the classification by the classification step based on the degree of matching with the predetermined distribution, wherein the evaluation value calculation step obtains an optimum evaluation by the classification evaluation step. The change evaluation value is calculated based on the mixture distribution having the classification.

An image monitoring method according to an eleventh aspect of the present invention includes a distribution type specifying step of specifying the type of distribution of the plurality of pixel values to one of a plurality of distribution types assumed in advance. The change evaluation value calculation step and the non-change value calculation step recognize the distribution type specified by the distribution specifying step as a type of distribution constituted by the set of normal values, and the change evaluation value or the non-change value It calculates the evaluation value.

According to a twelfth aspect of the present invention, in the image monitoring method, the change evaluation value calculating step and the non-change value calculating step each include a predetermined criterion for enabling superiority comparison between models having different numbers of free parameters. A method of calculating a value, wherein the change evaluation value or the non-change evaluation value is calculated based on the plurality of pixel values.

[0020]

Embodiments of the present invention will be described below with reference to the drawings. In the drawings, common elements are denoted by the same reference numerals, and redundant description is omitted.

Embodiment 1 FIG. 1 shows a system configuration diagram of the image monitoring apparatus according to the first embodiment of the present invention. The image monitoring device according to the present embodiment includes an image sensor 10. The image sensor 10 is a sensor that acquires a two-dimensional image data corresponding to a predetermined monitoring area and the monitoring area. The image sensor 10 is realized by, for example, a visible imaging device such as a television camera, an IR imaging device, or a synthetic aperture radar (SAR).

The image data obtained by the image sensor 10 can be understood as a set of a plurality of pixel values arranged two-dimensionally. Further, in this case, it can be understood that each of the plurality of pixel values corresponds to the state of a specific portion of the monitoring area having a two-dimensional spread. More specifically, each pixel value is calculated based on the intensity of visible light, the intensity of infrared light, or the intensity of a reflected wave caused by a radar transmission wave, which reaches the image sensor 10 from a specific portion corresponding to the pixel value. Equivalent to.

The image monitoring apparatus according to the present embodiment has an area dividing unit 1
2 is provided. The image data acquired by the image sensor 10 is supplied to the area dividing unit 12. The area dividing unit 12
A process of dividing image data corresponding to the entire monitoring area into area data corresponding to a plurality of predetermined areas is executed.

FIG. 2 shows image data 14 obtained by the image sensor 10 and a part 16, 18, and 20 of the area data generated by the area dividing unit 12. In the present embodiment, the image sensor 10 monitors a vast area including mountains, rivers, and forests. Therefore, the image data 14 has regions corresponding to mountains, rivers, and forests, respectively.

Of the plurality of pixel values constituting the image data 14, the pixel values corresponding to the mountain region indicate values that are close to each other. Similarly, pixel values corresponding to the river region and pixel values corresponding to the forest region indicate values that are close to each other. As described above, in the image data 14, there is a small area in which pixel values indicate values that are close to each other. The area dividing unit 12 outputs the image data 14
Is divided into small areas containing only pixel values that are particularly close to each other, and area data corresponding to each small area is generated. For this reason, the area data generated by the area dividing unit 12 includes only pixel values having approximate values unless a change occurs in the area.

FIG. 3 shows an example 22 of the area data generated by the area dividing section 12. Area data 2 shown in FIG.
No. 2 has a region having a higher pixel value than the other portions at substantially the center of the corresponding small region. FIG. 4 shows a distribution of pixel values included in the area data 22 shown in FIG. Note that the distribution shown in FIG. 4 is represented assuming that the pixel value included in the region data 22 most, that is, the pixel value that becomes the peak of the distribution is “0”.

In the present embodiment, the monitoring area of the image sensor 10 is set to a predetermined known area. Therefore, each of the area data generated by the area dividing unit 12 also corresponds to a preset known small area. In addition, the type of distribution of pixel values corresponding to a known small area is maintained substantially constant unless a change occurs in the small area. Hereinafter, this distribution shape is referred to as a basic distribution type. In the present embodiment, the image monitoring device stores basic distribution types corresponding to all small regions, as described later.

The small area corresponding to the area data 22 is an area whose basic distribution type is a normal distribution. Therefore, when no change occurs in the small area, the distribution of the area data 22 substantially follows the normal distribution. As described above, the area data 22 shown in FIG. 3 includes an area having a higher pixel value than other areas. For this reason, in the distribution illustrated in FIG. 4, a portion deviating from the normal distribution is formed in a region where the pixel value is large.

The image monitoring apparatus of this embodiment includes a change evaluation value calculation section 24 and a non-change evaluation value calculation section 26. The area data generated by the area dividing unit 12 is supplied to both the change evaluation value calculation unit 24 and the non-change evaluation value calculation unit 26. The change evaluation value calculation unit 24 calculates a change evaluation value J indicating the likelihood that a change has occurred in the area data.
This is the part for calculating A. On the other hand, the non-change evaluation value calculation unit 2
Reference numeral 6 denotes a part for calculating a non-change evaluation value JB indicating the certainty that no change has occurred in the area data.

FIG. 5 shows a flowchart of a series of processes executed by the change evaluation value calculating section 24. A series of processing shown in FIG. 5 is executed for each area data. FIG.
In the series of processes shown in (1), first, the process of step 30 is executed.

In step 30, a process of taking in area data from the area dividing section 12 is executed.

In step 32, assuming that a change has occurred in the region data, a process of modeling the distribution of pixel values included in the region data (hereinafter, referred to as pixel value distribution) is executed. Hereinafter, the model set in step 32 is referred to as a change assumption model. The change assumption model includes a pixel value distribution in which the pixel value distribution does not change from the normal value (hereinafter, referred to as a normal value distribution) and a pixel value distribution in which the pixel value changes from the normal value (hereinafter, the change value distribution). ) Is well suited to the actual pixel value distribution.

As described above, in the present embodiment, the basic distribution type corresponding to each small area is set in advance. The basic distribution type is a type of distribution formed by a set of normal pixel values. Therefore, also in the change assumption model, the type of the normal value distribution matches the basic distribution type. Further, in the present embodiment, a set of pixel values that have changed from the normal value due to the change of the small area can be treated as if they follow a normal distribution. Therefore, in the change assumption model, the type of the change value distribution can be regarded as a normal distribution.

As described above, in the present embodiment, both types of the two distributions (normal value distribution and change value distribution) constituting the change assumption model can be handled as known ones. A plurality of pixel values included in the area data are represented by 2
There are many classifications that can be divided into sets. Step 32 above
Then, among these many classifications, the classification that best matches the combination of the type of the known normal value distribution and the type of the known change value distribution is detected, and the mixture distribution according to the classification is set as the change assumption model. .

The processing in step 32 can be realized by using, for example, a known k-means method or a known robust clustering method. In the K-means method, a plurality of data included in a predetermined distribution are appropriately classified into two sets, and a likelihood and a log likelihood (an evaluation value indicating a degree of conformity to a known distribution type) for the classification are calculated. This is a method of detecting an optimal classification by repeating the above processing until the best classification is obtained based on likelihood and log likelihood.

On the other hand, the robust clustering has almost k-
The same process as the means method is performed, but when data is classified into two sets and an evaluation value for the classification is calculated, data far away from the center of each set is ignored and calculated.

In the present embodiment, in step 32, specifically, a process of detecting an optimal classification by the k-means method using log likelihood as an evaluation value is executed. Hereinafter, an example of the processing executed in step 32 will be described. In the following example, the number of pixels included in the region data is N, and those pixel values are scalar values yi (1 ≦ 1).
i ≦ N), and both the normal value distribution and the change value distribution follow a normal distribution.

In step 32, N pixel values yi (1 ≦ i) are set using an appropriate pixel value as a threshold value.
≦ N) is classified into a set 1 and a set 2. In the change evaluation value calculation unit 24, λik (1 ≦ i ≦ N; k = 1 or 2) is defined corresponding to the pixel number i and the set number k. When the pixel value yi is included in the set 1, λi1 and λi2 are λi1 = 1 and λi2 =
Set to 0. On the other hand, when the pixel value yi is included in the set 2, λi1 and λi2 are λi1 = 0 and λi2 =
Set to 1. According to the above λik, the classification result of the N pixel values yi can be represented by a matrix Λ of N rows and 2 columns.

The log likelihood J (Λ) of the matrix described above can be expressed by the following equation (1).

[0040]

(Equation 1)

Where yavek shown in the above equation (1) is a set k
Is the average value of the pixel values yi included in. Further, σk ² shown in the above equation (1) is the variance of the error of the pixel value yi included in the set k. These values are calculated by the following equations (2), respectively.
And (3).

[0042]

(Equation 2)

The variance σk ² of the error of the pixel value yi included in the set k may be calculated by the following equation (4) instead of the above equation (3).

[0044]

(Equation 3)

The log likelihood J calculated by the above equation (1)
(Λ) is a distribution of pixel values included in the set 1, and
The characteristic value is such that the distribution of pixel values included in the set 2 becomes smaller as the distribution conforms to the normal distribution. In the above step 32, the process of setting a different classification and calculating the log likelihood J (Λ) for the classification is repeated until a classification that minimizes the log likelihood J (Λ) is detected. as a result,
According to the processing of step 32, it is possible to obtain a change assumption model that is optimal for adapting both the normal value distribution and the change value distribution to a known distribution type.

The area data 22 shown in FIGS.
With respect to, the pixel value “6” is set as a threshold, pixel values less than the value “6” are classified into a set of normal values, and pixel values “6” or more are classified into a set of change values. In this case, the log likelihood J (Λ) can be minimized. Therefore, according to the process of step 32, the mixture distribution according to the above classification is set as the change assumed distribution for the region data 22. When the process of step 32 is completed, the process of step 34 is next executed in the change evaluation value calculation unit 24.

In step 34, a change evaluation value JA is calculated which indicates the certainty of assuming that the area data output from the area dividing section 12 is a mixture distribution of a normal value distribution and a change value distribution.

By the way, the image monitoring apparatus of the present embodiment
Based on a plurality of pixel values included in the area data, it is determined whether or not a change has occurred in the small area corresponding to the area data. Whether or not a change has occurred in the area data can be determined based on, for example, whether or not a pixel value different from the normal value is included in the area data. Whether or not a change has occurred in the region data can also be determined based on, for example, whether or not the pixel value distribution is different from the normal distribution.

However, the normal value of each pixel value, and
The normal distribution of pixel values included in the area data shows a large change due to external factors such as weather and season. For this reason, by setting the normal value or the normal distribution to a constant value or a constant distribution and performing the above determination, it is not always possible to accurately detect the occurrence of a change in the area data.

The image monitoring apparatus according to the present embodiment sets the change assumption model as described above as a model of the pixel value distribution, and, as described later, uses the non-change assumption based on the assumption that no change has occurred in the area data. Set the model. The non-change assumption model is a model having the same type of distribution as the basic distribution type.

The type of pixel value distribution (for example, normal distribution, Weibull distribution, K distribution, etc.) is almost always kept constant unless a small area changes. Therefore, when no change occurs in the small area, the type of the pixel value distribution should match the basic distribution type. In this case, the actual pixel value distribution better fits the non-change assumption model than the change assumption model.

On the other hand, when a change occurs in the small area and as a result, a change occurs in some pixel values included in the area data, the type of the actual pixel value distribution differs from the basic distribution type. . Then, when a clear change occurs in the small area, a situation occurs in which the actual pixel value distribution better fits the change assumption model than the non-change assumption model.

In other words, in the image monitoring apparatus of this embodiment, when the actual pixel value distribution conforms to the non-change assumption model rather than the change assumption model, it is determined that no change has occurred in the area data. Can be. If the actual pixel value distribution conforms to the change assumption model rather than the non-change assumption model, it can be determined that a change has occurred in the region data. According to such a method, it is possible to always determine the presence or absence of a change in a small area with excellent accuracy regardless of the weather or the season. The image monitoring apparatus according to the present embodiment is characterized in that the presence or absence of a change in a small area is determined by the above method.

As described above, the likelihood and the log likelihood are characteristic values indicating the degree of matching between the set model and the actual distribution. Therefore, according to the likelihood and the log likelihood, the suitability of the set model can be determined. However, the likelihood and the log likelihood have a tendency that the larger the number of free parameters of the model, the more likely it is to indicate a high degree of conformity. For this reason, it is not possible to compare the superiority between models having different numbers of free parameters depending on likelihood and log likelihood.

The change assumption model set in this embodiment includes two distributions. On the other hand, the non-change assumption model includes only one distribution. For this reason,
The numbers of free parameters of these two models are different from each other. Therefore, the superiority of the change assumption model and the non-change assumption model cannot be directly compared depending on the likelihood or the log likelihood.

As a method for enabling superiority comparison between models having different numbers of free parameters, for example, AIC (Ak
aike Information Criterion), MDL (Minimum Descripti
On Length) standards and techniques using the MDL principle and the like are known. The image monitoring device according to the present embodiment compares the superiority of the change assumption model and the superiority of the non-change assumption model by using the MDL principle among these methods.

That is, in step 34 described above, MD
The change evaluation value JA for the change assumption model is calculated by a method based on the L principle. The change evaluation value JA for the change assumption model set by the process of step 32 is:
It can be obtained by the following equation (5).

[0058]

(Equation 4)

The change evaluation value J obtained by the above equation (5)
A is a characteristic value that becomes smaller as the change assumption model conforms to the actual pixel value distribution. When the above calculation processing is completed, the change evaluation value calculation unit 24 executes the processing of step 36 next. In step 36,
A process for specifying the change assumption range C is executed. In this step 36, the existence range of the pixel values classified into the change value set when the change assumption model is set is specified as the range change assumption range C.

FIG. 6 shows the area data 22 shown in FIG.
With the pixel value “6” or more and the pixel value “6”
The figure which classify | categorized and represented to the part below is shown. Area data 2
2 is the target of processing, in this step 36, FIG.
The range outlined in white is specified as the change assumed range C. When the above processing is completed, the change evaluation value calculation unit 24
Is completed.

FIG. 7 shows a flowchart of a series of processes executed by the non-change evaluation value calculating section 26. A series of processing shown in FIG. 7 is executed for each area data. In the series of processes shown in FIG. 7, first, the process of step 40 is executed.

In step 40, a process of taking in the area data from the area dividing section 12 is executed.

In step 42, assuming that no change has occurred in the area data, a process of modeling the pixel value distribution with the same kind of distribution as the basic distribution type is executed. Hereinafter, the model set in step 42 is referred to as a non-change assumption model. The non-change assumption model fits well with the actual pixel value distribution when all pixel values included in the region data are included in the set of normal values.

In step 44, a non-change evaluation value JB is calculated which indicates the certainty that the area data output from the area dividing section 12 is assumed to be a distribution composed of only normal value distributions. In the present embodiment, the non-change evaluation value JB is calculated by a method according to the MDL principle, similarly to the above-described change evaluation value JA. The non-change evaluation value JB is, for example,
When the basic distribution type of the area data is the normal distribution and the area data includes the pixel value yi (1 ≦ i ≦ N) of the number N of pixels, it is expressed by the following equation (6).

[0065]

(Equation 5)

Here, yave shown in the above equation (6) is an average value of all pixel values yi. Further, σ2 shown in the above equation (6) is the variance of the error of all the pixel values yi. These values can be expressed by the following equations (7) and (8), respectively.

[0067]

(Equation 6)

The non-change evaluation value JB expressed by the above (6)
Are characteristic values that become smaller as the non-change assumption model matches the actual pixel value distribution. The above-mentioned non-change evaluation value JB is compared with a change evaluation value JA (refer to the above equation (5)) calculated for the same area data, thereby comparing the non-change assumption model with the change assumption model. Is a value that enables In the above step 44, the non-change evaluation value JB
Is calculated, a series of processing to be executed by the non-change evaluation value calculation unit 26 ends.

As shown in FIG. 1, the image monitoring apparatus of this embodiment includes a change judging section 48. The change evaluation value JA and the change assumption range C calculated or specified by the change evaluation value calculation unit 24 and the non-change evaluation value JB calculated by the non-change evaluation value calculation unit 26 are supplied to the change determination unit 48. . The change determination unit 48 determines whether there is a change in the small area and specifies a change range by executing a process described later based on the various data supplied as described above.

FIG. 8 shows a flowchart of a series of processing executed by the change judging section 48. A series of processes shown in FIG. 8 is executed for each area data. In the series of processes shown in FIG. 8, first, the process of step 50 is executed.

In steps 50 to 54, the change evaluation value JA, the change assumption range C, and the non-change evaluation value JB
Is performed.

In step 56, a process for comparing the change evaluation value JA with the non-change evaluation value JB is executed. As described above, the change evaluation value JA and the non-change evaluation value JB are calculated according to the MDL principle, and become smaller as the model fits the actual distribution. Therefore, when JA <JB holds, it can be determined that the change assumption model is superior to the non-change assumption model. On the other hand, when JA> JB holds,
It can be determined that the non-change assumption model is superior to the change assumption model.

In step 58, the comparison processing in step 56 indicates that the change assumption model is superior to the non-change assumption model, that is, the change assumption model is appropriate as the pixel value distribution model. Is determined. As a result, when it is determined that the change assumption model is appropriate, it is appropriate to determine that some change has occurred in the region data. In this case, the process of step 60 is executed after step 58. On the other hand, if it is determined in step 58 that the change assumption model is not appropriate, it is appropriate to determine that no change has occurred in the region data. In this case, the process of step 62 is executed next.

At step 60, a judgment result R1 indicating the occurrence of a change in the area data and a change range CR in the area data are output. The change range CR is a range that matches the assumed change range C supplied from the change evaluation value calculation unit 24.

At step 62, a judgment result R2 indicating that no change has occurred in the area data is output. When the processing in step 62 or the processing in step 60 ends, a series of processing to be executed in the change determination unit 48 ends.

As shown in FIG. 1, the image monitoring apparatus according to the present embodiment includes a judgment result collecting unit 64. Change judgment unit 4
The judgment results R1 and R2 and the change range CR output by 8 are supplied to the change judgment result collection unit 64. The change determination result collection unit 64 collects the determination results R1 and R2 and the change range CR output corresponding to the individual region data, and generates data corresponding to the entire monitoring region of the image sensor 10.
Therefore, according to the data generated by the change determination result collection unit 64, it can be determined whether or not a change has occurred in the entire monitoring area.

As described above, according to the image monitoring apparatus of the present embodiment, the presence or absence of a change in the monitoring area is determined based on which of the change assumption model and the non-change assumption model is superior as the pixel value distribution model. can do. The superiority of the change assumption model and the non-change assumption model is determined only by whether or not a change occurs in the monitoring area without being affected by external factors such as weather and season. For this reason, according to the image monitoring device of the present embodiment, the presence or absence of a change in the monitoring area can always be determined with excellent accuracy.

In the above embodiment, the pixel value is limited to a scalar value. However, the present invention is not limited to this. A pixel value is defined as a vector including a plurality of elements such as intensity and phase. Alternatively, it may be represented by an imaginary number or the like.

In the above embodiment, the change evaluation value calculation unit 24 uses the k-means method as a method of classifying a plurality of pixel values into two sets, but the present invention is not limited to this. Instead, robust clustering may be used as the above method.

Further, in the above embodiment, a method based on the MDL principle is used as a method for judging the superiority of the change assumption model and the non-change assumption model, but the present invention is not limited to this. Instead, the above determination may be made by, for example, a method using AIC or MDL standards.

Further, in the above embodiment, the change evaluation value calculating section 24 changes the pixel value classified into the change value set, that is, changes the existence range difference of the pixel value having a value equal to or more than the predetermined threshold value. Although the method is specified as the assumed range C, the method of specifying the assumed change range C is not limited to this, and the change evaluation value calculation unit 24 may use the pixel values of the pixel values classified into the smaller set of pixels. The existence range may be specified as a change assumption range C.

Further, in the above embodiment, the pixel values constituting the image data are directly supplied to the area dividing section 12, but the present invention is not limited to this, and some filter processing is performed. The subsequent pixel value may be supplied to the area dividing unit 12.

For example, when the number of pixel values constituting the image data is large,
It is effective to execute a process of converting a plurality of adjacent pixel values into one pixel value. According to the above processing, the number of pixel values supplied to the area dividing unit 12 can be reduced. Therefore, according to the above-described processing, when a large number of pixel values are included in the image data, a desired monitoring function can be realized without performing an unnecessarily large amount of processing. .

For example, when a large noise is superimposed on image data, a moving average process for converting one pixel value into an average value of a plurality of adjacent pixel values may be executed as the above filter process. It is valid. According to the above processing, the influence of noise on individual pixel values can be effectively suppressed. For this reason, by executing the above-described processing, it is possible to realize an accurate monitoring function in a situation where noise is likely to be superimposed on image data.

In the above embodiment, the "distribution detecting means" and "change evaluation value calculating means" according to the first aspect and the "distribution detecting means" according to the twelfth aspect are obtained by the processing of the steps 30 to 34. 2. The "non-change evaluation value calculation means" according to claim 1, wherein the "detection step" and the "change evaluation value calculation step" are performed by the processing of steps 40 to 44.
And the “non-change evaluation calculating step” according to claim 12 are realized, respectively.

In the above embodiment, the "classification means" according to the third aspect and the "classification step" according to the ninth aspect can be replaced by the processing in the steps 36, 58 and 60 by the processing in the step 32. Accordingly, the "change range specifying means" according to claim 3 and the "change range specifying step" according to claim 9 are realized respectively.

In the above embodiment, the "classification changing means" and the "classification evaluation means" according to the fourth aspect and the first aspect are executed by the processing of the step 32.
The “classification changing step” and “classification evaluation step” described in No. 0 are realized, respectively.

Further, in the above embodiment, AIC, M
A method using the DL standard or the MDL principle corresponds to the "method of calculating a predetermined reference value enabling comparison between models having different numbers of free parameters".

Embodiment 2 Next, an image monitoring apparatus according to a second embodiment of the present invention will be described with reference to FIG.
FIG. 9 shows a system configuration diagram of the image monitoring apparatus of the present embodiment. The image monitoring apparatus according to the present embodiment is characterized in that a reference image storage unit 70 and a difference image creation unit 72 are provided in the system configuration shown in FIG.

The reference image storage 70 stores image data representing the reference state of the monitoring area as a reference image. The reference image storage unit 70 supplies the above-described reference image to the difference image creation unit 72. The difference image creation unit 72 is supplied with the reference image from the reference image storage unit 70 and the image data from the image sensor 10. The difference image creation unit 72 creates a difference image between the image data and the reference image by performing processing described later, and supplies the created difference image to the region division unit 12.

FIG. 10 shows a flowchart of a series of processing executed in the difference image creating section 72. The image monitoring apparatus according to the present embodiment executes a monitoring process of a monitoring area at predetermined time intervals, for example, every hour. A series of processes shown in FIG. 10 is repeatedly executed at every predetermined time. FIG.
In the process shown in (1), first, the process of step 80 is executed.

In step 80, a process for taking in image data from the image sensor 10 is executed. This step 80
The image data captured in step (1) is treated as a set of two-dimensionally arranged pixel values in the difference image creation unit 72.

At step 82, a process of taking in the reference image from the reference image storage unit 70 is executed. Step 8
The reference image captured in 2 is treated as a set of two-dimensionally arranged pixel values in the difference image creation unit 72, similarly to the image data.

In step 84, a process of generating a difference image between the image data and the reference image by obtaining a difference between corresponding pixel values and outputting the difference image to the area dividing unit 12 is executed.

In step 86, the reference image storage unit 70
Then, a process of storing the image data acquired in the current processing cycle as a new reference image is executed. When the processing in step 86 ends, a series of processing to be executed in the difference image creating unit 72 ends.

According to the above-described processing, the difference image between the image data and the reference image 70 is supplied to the area dividing section 12. The pixel values constituting the difference image are “0” in a range where the image data and the reference image match, and are different from “0” in a range where the image data and the reference image do not match.

The image data corresponds to a wide monitoring area. For this reason, in the image data, a region having a relatively large pixel value (normal value) and a region having a relatively small pixel value are mixed. For this reason, if the area data is created directly from the image data by dividing the image data, a bias error due to a difference in position in the image data remains in the created plurality of area data.

On the other hand, the above-described difference image has a pixel value “0” substantially equal in the whole area, though it corresponds to a wide monitoring area. For this reason,
By creating the area data by dividing the difference image, it is possible to obtain area data without a bias error. Further, according to the region data created by the above-described method, it is possible to simply recognize a portion having a large absolute value of the pixel value as a changed portion. Therefore, according to the image monitoring device of the present embodiment, a highly accurate monitoring function can be easily realized as compared with the image monitoring device of the first embodiment.

In the above embodiment, the monitoring process by the image monitoring device is executed at predetermined time intervals. However, the present invention is not limited to this. Alternatively, only the reference image update process may be executed at predetermined time intervals.

In the above embodiment, the "reference image data storage means" described in claim 2 is changed by the reference image storage means 70 to the "reference image data storage means" described in claim 8 by the processing in step 86. The "step" implements the "difference image creation means" according to claim 2 and the "difference image creation step" according to claim 8 by the processing of step 84, respectively.

Embodiment 3 FIG. Next, an image monitoring apparatus according to a third embodiment of the present invention will be described with reference to FIGS. FIG. 11 shows a system configuration diagram of the image monitoring apparatus of the present embodiment. The image monitoring apparatus of the present embodiment includes a distribution test unit 90. The distribution test unit 90 is supplied with region data from the region division unit 12. The distribution test unit 90 tests the type of distribution of the pixel values included in the region data by performing the processing described below, and specifies the type of distribution as one of a plurality of preset distribution types.

The image monitoring apparatus of this embodiment includes a distribution test type change evaluation value calculation unit 92 and a distribution test type non-change evaluation value calculation unit 94. Distribution test type change evaluation value calculation unit 9
2 and the distribution test type non-change evaluation value calculation unit 94 are supplied with the region data from the region division unit 12 and the data on the type of distribution specified by the distribution test unit 90.

In the image monitoring apparatuses of the first and second embodiments, the change evaluation value JA and the non-change evaluation value JB
Is calculated on the assumption that the basic distribution type (the distribution type when there is no change) of the pixel values included in the region data is known. Therefore, in the image monitoring apparatuses of the first and second embodiments, it is necessary to determine the basic distribution type in advance for all the small areas included in all the monitoring areas.

In the image monitoring apparatus according to the present embodiment, the distribution test type change evaluation value calculation section 92 and the distribution test type non-change evaluation value calculation section 94 determine whether the basic distribution type of the pixel values included in the area data is the distribution test section 90. It is characterized in that the change evaluation value JA and the non-change evaluation value JB are calculated as the distribution type specified by. According to the above method, the change evaluation value JA and the non-change evaluation value JB can be obtained with high accuracy for all the small regions without previously determining the basic distribution type for the small regions.

FIG. 12 is a flowchart of a series of processes executed in the distribution test unit 90. A series of processing shown in FIG. 12 is executed for each area data. In the process shown in FIG. 12, first, the process of step 100 is executed.

In step 100, a process of taking in area data from the area dividing section 12 is executed.

In step 102, the type of distribution of pixel values included in the area data is tested. In the present embodiment, the small area corresponding to each area data is set so that the range where the change occurs is sufficiently smaller than the range where the change does not occur. For this reason, the distribution type of the pixel values included in the area data can always be regarded as the same type regardless of the presence or absence of a change in the area data. In the present embodiment, the distribution type is referred to as a basic distribution type.

The image monitoring apparatus according to the present embodiment stores distribution types such as a normal distribution, a Weibull distribution, and a K distribution in advance. In step 102, specifically, a process of selecting a distribution type that most closely approximates the actual pixel value distribution from among those distribution types stored in advance by a technique such as a fitness test is executed.

In step 104, the above step 102
The distribution type selected in is identified as the basic distribution type,
Further, data representing the basic distribution type is output to the distribution test type change evaluation value calculation unit 92 and the distribution test type non-change evaluation value calculation unit 94. When the processing of this step ends, a series of processing to be executed by the distribution testing unit 90 ends.

In the image monitoring apparatus of this embodiment, the distribution test type change evaluation value calculation unit 92 assumes that, of the pixels included in the area data, the pixel values included in the normal value set follow the basic distribution type described above. A model is set, and a change evaluation value JA is calculated by the same processing as in the first embodiment. In the first embodiment,
Although it is assumed that the distribution type of the pixel values included in the change value set is known, in the present embodiment, the distribution type may be known or may be the same as the basic distribution type.

Further, in the image monitoring apparatus of the present embodiment, the distribution test type non-change evaluation value calculating section 94 determines the non-change assumption model on the assumption that the pixel values included in the region data form the same kind of distribution as the basic distribution type. Then, the non-change evaluation value JB is calculated by the same processing as in the first embodiment. According to the above-described processing, it is not necessary to determine the basic distribution type for the small area in advance, and it is possible to accurately reflect the type of the actual pixel value distribution on the change evaluation value JA and the non-change evaluation value JB. Become. For this reason,
According to the image monitoring apparatus of the present embodiment, it is possible to easily and accurately detect a change in the monitoring area.

In the above embodiment, the “distribution type specifying means” according to the fifth aspect and the “distribution type specifying step” according to the eleventh aspect can be implemented by a distribution test type Change evaluation value calculation unit 92
The “change evaluation value calculation means and non-change value calculation means” according to claim 5 and the “change evaluation value calculation step and non-change value” according to claim 11, by the distribution test type non-change evaluation value calculation unit 94. Calculation step "is realized respectively.

[0113]

Since the present invention is configured as described above, it has the following effects. Claim 1
According to the invention described in the seventh aspect, by comparing the change evaluation value with the non-change evaluation value, it is possible to appropriately determine whether the pixel value distribution should be assumed to be a mixture distribution or a non-change distribution. Can be. Then, it is possible to determine whether or not a change has occurred in the monitoring area based on the determination result. According to the above method, a change in the monitoring area can always be detected with excellent accuracy without being affected by a change in weather, season, or the like.

According to the second or eighth aspect of the present invention, the distribution detecting means can detect the distribution of a plurality of pixel values constituting the difference image. The image data and the reference image data are almost equally affected by external factors such as weather. Therefore, the difference image data is hardly affected by these external factors. In the present invention, the presence or absence of a change in the monitoring area is determined based on the difference image data having the above characteristics. According to such a method, a change in the monitoring area can always be detected with high accuracy without being affected by external factors.

According to the third or ninth aspect of the present invention, when it is determined that a change has occurred in the monitoring area, a set of parts corresponding to pixel values belonging to the change value set is recognized as a change range. According to the above-described method, the range that has changed from the normal state in the monitoring area can be easily and accurately detected.

According to the fourth or tenth aspect of the present invention,
A classification that best matches the distribution of pixel values belonging to the normal value set and the distribution of pixel values belonging to the change value set with a predetermined distribution is specified, and a change evaluation value is calculated based on a mixture distribution according to the classification. According to the above method, the certainty of assuming that the distribution of pixel values is a mixture distribution can be determined based on the optimal mixture distribution. Therefore, according to the present invention, it is possible to accurately determine whether there is a change in the monitoring area.

According to the invention described in claim 5 or 11,
The change evaluation value and the non-change evaluation value can be calculated after specifying the type of distribution constituted by a set of normal values as an appropriate distribution. These evaluation values can be calculated with higher accuracy as the assumptions set at the time of the calculation match the actual state. According to the above processing, the type of distribution constituted by the set of normal values can be set to a type close to the actual distribution type. Therefore, according to the present invention, it is possible to accurately determine whether there is a change in the monitoring area.

According to the invention of claim 6 or 12,
Each of the change evaluation value and the non-change evaluation value is calculated as a reference value that enables superiority comparison between models having different numbers of free parameters. The change evaluation value is calculated assuming a model in which a plurality of pixel values constitute two distributions. On the other hand, the non-change evaluation value is calculated assuming a model in which a plurality of pixel values constitute one distribution. These two
One model contains a different number of free parameters. For this reason, in order to judge the superiority of these two models, it is necessary to judge in consideration of the difference in the number of free parameters. According to the present invention, superiority comparison that satisfies the above requirements is performed. Therefore, according to the present invention, it is possible to accurately determine whether there is a change in the monitoring area.

[Brief description of the drawings]

FIG. 1 is a system configuration diagram of an image monitoring device according to a first embodiment of the present invention.

FIG. 2 is an example of image data obtained by the image monitoring device shown in FIG.

FIG. 3 is an example of area data obtained by the image monitoring device shown in FIG. 1;

FIG. 4 is a distribution of pixel values included in the area data shown in FIG. 3;

FIG. 5 is a flowchart of a series of processing executed by a change evaluation value calculation unit shown in FIG. 1;

FIG. 6 is a diagram illustrating a change assumption range specified by the routine shown in FIG. 5;

FIG. 7 is a flowchart of a series of processes executed by a non-change evaluation value calculation unit shown in FIG. 1;

FIG. 8 is a flowchart of a series of processes executed by a change determination unit shown in FIG. 1;

FIG. 9 is a system configuration diagram of an image monitoring device according to a second embodiment of the present invention.

FIG. 10 is a flowchart of a series of processing executed by a difference image creating unit shown in FIG. 9;

FIG. 11 is a system configuration diagram of an image monitoring device according to a third embodiment of the present invention.

FIG. 12 is a flowchart of a series of processes executed by a difference image creating unit shown in FIG. 11;

[Explanation of symbols]

Reference Signs List 10 image sensor, 12 area dividing section, 14 image data, 16 to 22 area data, 24 change evaluation value calculation section, 26 non-change evaluation value calculation section, 48 change judgment section,
70 reference image storage unit, 72 difference image creation unit, 90
Distribution tester, 92 Distribution test type change evaluation value calculator, 94
Distribution test type non-change evaluation value calculation section, JA change evaluation value,
JB Non-change evaluation value, C change assumed range, R1, R2 judgment result, CR change range.

Claims (12)

[Claims] An image sensor capable of outputting a plurality of pixel values constituting image data of a monitoring area; a distribution detecting unit detecting a distribution of the plurality of pixel values; A set and a change evaluation value calculating means for calculating a change evaluation value representing a probability of assuming that the mixture distribution is a set of change values changed from the normal value, and a distribution of the pixel values, A non-change evaluation value calculating means for calculating a non-change evaluation value representing a probability of assuming a non-change distribution composed of only a set of values, and the change evaluation value, based on the non-change evaluation value Determining means for determining whether a change from a normal state has occurred in the monitoring area. 2. A reference image data storage unit for storing reference image data corresponding to a reference state of the monitoring area; and a difference image corresponding to a difference between the image data acquired by the image sensor and the reference image data. The image monitoring apparatus according to claim 1, further comprising: a difference image creating unit that creates the difference image, wherein the distribution detecting unit detects a distribution of a plurality of pixel values included in the difference image. 3. The change evaluation value calculating means classifies each of the plurality of pixel values into a normal value set assuming the set of normal values and a change value set assuming the set of change values. It is provided with a classifying unit, and when it is determined that a change has occurred in the monitoring area, a change range specifying unit that specifies a set of portions corresponding to pixel values belonging to the set of change values as a change range. The image monitoring device according to claim 1 or 2, wherein 4. A classification changing means for appropriately changing the classification by the classification means; a distribution of pixel values belonging to the set of normal values and a distribution of pixel values belonging to the set of change values; Classification evaluation means for evaluating the suitability of the classification by the classification means based on the degree of conformity, the evaluation value calculation means, based on the mixture distribution having the classification obtained the optimal evaluation by the classification evaluation means 4. The image monitoring apparatus according to claim 3, wherein a change evaluation value is calculated. 5. The type of distribution of the plurality of pixel values,
A distribution type identification unit that identifies one of a plurality of distribution types assumed in advance, and the change evaluation value calculation unit and the non-change value calculation unit include a distribution type identified by the distribution identification unit. 5. The image monitoring apparatus according to claim 4, wherein the change evaluation value or the non-change evaluation value is calculated by recognizing the type of distribution constituted by a set of normal values. 6. The method according to claim 1, wherein the change evaluation value calculation means and the non-change value calculation means calculate a predetermined reference value enabling comparison between models having different numbers of free parameters. The image monitoring apparatus according to claim 1, wherein the change evaluation value or the non-change evaluation value is calculated based on a value. 7. A pixel value detecting step of obtaining a plurality of pixel values constituting image data of a monitoring area from an image sensor; a distribution detecting step of detecting a distribution of the plurality of pixel values; and a distribution of the pixel values. A set of normal values, a change evaluation value calculating step of calculating a change evaluation value representing the probability of assuming that it is a mixture distribution composed of a set of change values changed from the normal value, A non-change evaluation value calculating step of calculating a non-change evaluation value representing a probability of assuming that the distribution is a non-change distribution composed of only the set of the normal values; the change evaluation value; and the non-change evaluation. A determination step of determining whether a change from a normal state has occurred in the monitoring area based on the value. 8. A reference image data storing step of storing reference image data corresponding to a reference state of the monitoring area, and a difference image corresponding to a difference between the image data acquired by the image sensor and the reference image data. The image monitoring method according to claim 7, further comprising a difference image creating step of creating, wherein the distribution detecting step detects a distribution of a plurality of pixel values constituting the difference image. 9. The change evaluation value calculating step classifies each of the plurality of pixel values into a normal value set assumed as the set of normal values and a change value set assumed as the set of change values. The method further comprises a classification step, and when it is determined that a change has occurred in the monitoring area, a change range specifying step of specifying a set of parts corresponding to pixel values belonging to the set of change values as a change range. The image monitoring method according to claim 7 or 8, wherein 10. A classification changing step of appropriately changing the classification in the classification step; a distribution of pixel values belonging to the set of normal values and a distribution of pixel values belonging to the set of changed values; A classification evaluation step of evaluating the suitability of classification by the classification step based on the degree of adaptation, wherein the evaluation value calculation step is based on a mixture distribution having a classification obtained by the classification evaluation step with an optimum evaluation. 10. A change evaluation value is calculated.
Image monitoring method as described. 11. A distribution type specifying step of specifying a type of distribution of the plurality of pixel values to one of a plurality of distribution types assumed in advance, and the change evaluation value calculating step and the non-change value The calculating step recognizes the distribution type specified by the distribution specifying step as a type of distribution constituted by the set of normal values, and calculates the change evaluation value or the non-change evaluation value. Item 10. The image monitoring method according to Item 10. 12. The change evaluation value calculation step and the non-change value calculation step are each a method of calculating a predetermined reference value enabling comparison between models having different numbers of free parameters. 12. The image monitoring method according to claim 7, wherein the change evaluation value or the non-change evaluation value is calculated based on a value.