JP3910626B2 - Monitoring device - Google Patents

Description

The present invention relates to a monitoring device that monitors a plurality of moving objects such as a moving person or a vehicle or a crowd.

  As a conventional monitoring device, a device shown in FIG. 17 is known (see Patent Document 1). As shown in FIG. 17, this conventional monitoring apparatus 10 includes an imaging device 11, a part position extraction unit 12, a shape feature quantity extraction unit 13, a part position tracking unit 14, a motion feature quantity measurement unit 15, a comparison search unit 16, and A display unit 17 is provided.

  As shown in FIG. 17, in the conventional monitoring device 10, the part position extraction unit 12 extracts a person's part position, the shape feature amount extraction unit 13 extracts a shape feature amount, and the part position tracking unit 14 extracts the shape feature amount. The movement feature amount measurement unit 15 tracks the movement feature amount. The conventional monitoring device 10 automatically extracts the feature amount of the person in the monitoring area, creates a database, searches for a candidate person having the characteristics of a specific suspicious person, and prompts the monitor to pay attention. Investigate, track and identify suspicious persons.

  As another conventional monitoring apparatus, there is an apparatus described in Patent Document 2. Other conventional monitoring devices capture the image of the face portion by automatically selecting and tracking the most appropriate tracking target by detecting the skin color and movement of the subject captured by the imaging device. Recognize an individual by analyzing the captured face image. Furthermore, it is determined as a dangerous person by identifying a mask, sunglasses, a hat, and the like from the information on the facial part.

However, in the conventional monitoring device, when a person is detected and collation with a database is performed, it is difficult to completely automate collation with a database accumulated in large quantities by the current technology. Further, in the conventional monitoring device, it is difficult to accurately specify the position of the part and extract the face image when there are images of a plurality of persons in the entire image or when occlusion occurs. . For this reason, if a person is detected and tracked and collation with a large number of databases is performed fully automatically, there are two concerns that a detection error, a tracking error, and a collation error will occur. There is also a danger of causing trouble.
JP 11-399139 A JP 2000-163600 A

  The present invention has been made in view of such a point, and by detecting a moving object or crowd image existing in the entire image in a macro rather than individually detecting and tracking, a part such as a human face is detected. An object of the present invention is to provide a monitoring device that can prevent a detection error, a tracking error, and a collation error even when it cannot be detected accurately, judge an abnormal situation, and call attention to a monitor.

  In order to achieve the above object, a monitoring apparatus according to the present invention is a monitoring apparatus that performs monitoring in a monitoring area, and that captures an image in the monitoring area and generates an entire image, and the imaging means A moving body image generating unit that generates a moving body image that is a plurality of moving objects or crowd images from the entire image generated by the moving body image generation unit, and the moving body image is generated based on the moving body image generated by the moving body image generating unit. A density calculation means for calculating a density indicating the degree of congestion of objects or the crowd, the density at the current time calculated by the density calculation means, and a reference density that is a density as a criterion for situation determination A panning / tilting / zooming machine, a situation judging means for judging that the current time situation is an abnormal situation when the density at the current time and the reference density differ from each other by a predetermined threshold or more The other image pickup means having at least one of the above and the other image pickup means to pick up an image of an area where the congestion at the current time is different from the reference congestion when the situation determination means And an imaging control means for controlling the imaging means.

  As described above, according to the monitoring device according to the present invention, a moving object or a partial image of a crowd existing in an image is not detected and tracked individually, but captured in a macro manner, so that the face of a person, etc. Even if the location of the object cannot be accurately detected, it is possible to prevent a detection error, a tracking error, and a verification error by judging an abnormal situation when the density or moving direction of the moving object or the crowd is significantly different from the normal time. You can call attention.

  A monitoring device according to an embodiment of the present invention is a monitoring device that performs monitoring in a monitoring region, and is generated by an imaging unit that captures an image in the monitoring region and generates an entire image, and the imaging unit. A moving body image generating unit that generates a moving body image that is an image of a plurality of moving objects or a crowd from the whole image, and the moving object or the crowd based on the moving body image generated by the moving body image generating unit And a density calculation means for calculating the density as a parameterized density by function approximation or frequency conversion.

  According to this configuration, even if a part such as a person's face cannot be accurately detected by capturing a macro object rather than individually detecting and tracking a moving object or crowd image existing in the entire image, the moving object can be moved. The density of objects or crowds can be calculated.

  Further, the monitoring device further determines the current time based on the congestion at the current time calculated by the congestion calculating means and a reference congestion that is a congestion as a reference for determining the situation. It is preferable to provide the situation judgment means to do.

  According to this configuration, it is possible to prevent a detection error, a tracking error, and a verification error by determining an abnormal situation when the density of the moving object or the crowd is significantly different from the normal time.

  Further, the monitoring device is further calculated by reference density calculation means for calculating the reference density based on the density at a predetermined date and time calculated by the density calculation means, and by the reference density calculation means. And a database that holds the reference density.

According to this configuration, it is possible to calculate the density of a moving object or a crowd at a normal time.
Further, the reference density calculation means calculates a new reference density every predetermined time based on the density calculated by the density calculation means, and calculates the reference density of the database with the calculated new reference density. The congestion level may be updated.

  According to this configuration, since the standard density of the database is updated every predetermined time, the determination of the abnormal situation becomes more accurate.

  In addition, the reference density calculation unit may calculate the reference density for at least one of every season, every time, every weather, or every day of the week, and store it in the database.

  According to this configuration, since the reference density for each season, every time, every weather, or every day of the week is calculated and stored in the database, the determination of an abnormal situation becomes more accurate.

  The density calculation unit and the reference density calculation unit may parameterize the density and the reference density by function approximation.

  Further, the density calculation means and the reference density calculation means may parameterize the density and the reference density by using the number of Gaussian distributions and the average value and variance value of each Gaussian distribution as parameters. Good.

  According to this configuration, by expressing the parameters, it is possible to determine an abnormal situation robustly against noise and detection errors.

  The monitoring apparatus may further include a database that holds the reference density set in advance.

  According to this configuration, the configuration can be simplified by setting and holding the reference density in advance.

  Further, the situation determination means compares the congestion at the current time with the reference congestion, and if the congestion at the current time and the reference congestion are different from each other by a predetermined threshold or more, the situation at the current time May be determined to be an abnormal situation.

  According to this configuration, it is possible to prevent a detection error, a tracking error, and a verification error by determining an abnormal situation when the density of the moving object or the crowd is significantly different from the normal time.

  The monitoring apparatus may further include a notification unit that notifies the situation at the current time determined by the situation determination unit.

  According to this configuration, it is possible to prevent a detection error, a tracking error, and a collation error by determining an abnormal situation when the density of the moving object or the crowd is significantly different from the normal time, and to alert the monitor.

  Further, the moving body image generating means may generate the moving body image by performing inter-frame difference processing or background difference processing on the entire image.

According to this configuration, it is possible to easily generate a moving body image.
The density calculation unit may detect a head candidate area from the whole image, and calculate the density based on an area occupied by the detected head candidate area with respect to the whole image.

According to this configuration, it is easy to calculate the density of moving objects or crowds.
The density calculation unit may calculate the density based on an area of the moving body image.

According to this configuration, it is easy to calculate the density of moving objects or crowds.
The density calculation unit may convert the moving body image into one-dimensional information and calculate the density based on the converted one-dimensional information.

According to this configuration, it is easy to calculate the density of moving objects or crowds.
The density calculation means may calculate the density by performing a frequency analysis of the one-dimensional information.

According to this configuration, it becomes easier to calculate the density of moving objects or crowds.
The imaging unit may be an infrared camera, and the density calculation unit may calculate the density based on a reaction area of the infrared camera.

According to this configuration, it is easy to calculate the density of moving objects or crowds.
The monitoring apparatus according to the embodiment of the present invention is a monitoring apparatus that performs monitoring in a monitoring area, and generates an entire image by capturing an image in the monitoring area, and is generated by the imaging means. A moving body image generating means for generating a moving body image which is a plurality of moving objects or crowd images from the whole image, and the moving object or the moving object based on the moving body image generated by the moving body image generating means. A crowd density calculating means for calculating a crowd density indicating the crowd density; the crowd density at the current time calculated by the crowd density calculating means; and a reference crowd density that is a crowd density as a criterion for situation determination; And the situation determination means for determining that the situation at the current time is abnormal when the density at the current time and the reference density are different from each other by a predetermined threshold or less, and the pan, tilt and zoom functions are reduced. The other imaging unit and the other imaging unit so as to capture an area where the density at the current time is different from the reference density when the situation determination unit determines that the situation is abnormal. Imaging control means for controlling the means.

  According to this configuration, it is possible to obtain a detailed image by causing the other imaging unit to image the occurrence region of the abnormal situation based on the determination result of the abnormal situation.

  The monitoring apparatus according to the embodiment of the present invention is a monitoring apparatus that performs monitoring in a monitoring area, and generates an entire image by capturing an image in the monitoring area, and is generated by the imaging means. A moving body image generating means for generating a moving body image which is a plurality of moving objects or crowd images from the whole image, and the moving object or the moving object based on the moving body image generated by the moving body image generating means. Based on the density calculation means for calculating the density indicating the degree of crowd density, the whole image generated by the photographing means and the density calculated by the density calculation means, The moving direction calculating means for calculating the moving direction of the crowd is provided.

  According to this configuration, even if a part such as a person's face cannot be accurately detected by capturing a macro object rather than individually detecting and tracking a moving object or crowd image existing in the entire image, the moving object can be moved. The direction of movement of the object or crowd can be calculated.

  In addition, the monitoring device further determines the situation at the current time based on the movement direction at the current time calculated by the movement direction calculation means and a reference movement direction that is a movement direction as a reference for situation determination. And a situation determination means for performing the above.

  According to this configuration, an abnormal situation can be determined when the moving direction of the moving object or the crowd is significantly different from the normal direction.

  Further, the monitoring device is further calculated by a reference movement direction calculation unit that calculates the reference movement direction based on the movement direction at a predetermined date and time calculated by the movement direction calculation unit, and the reference movement direction calculation unit. And a database that holds the reference movement direction.

According to this configuration, it is possible to calculate the normal movement direction.
Further, the reference movement direction calculation means calculates a new reference movement direction every predetermined time based on the movement direction calculated by the movement direction calculation means, and uses the calculated new reference movement direction for the reference of the database. The moving direction may be updated.

  According to this configuration, since the reference movement direction of the database is updated every predetermined time, the determination of the abnormal situation becomes more accurate.

  In addition, the reference movement direction calculation means may calculate the reference movement direction for each season, every time, every weather, or every day of the week, and store it in the database.

  According to this configuration, since the reference movement direction for each season, every time, every weather, or every day of the week is calculated and given to the database and held in the database, the determination of the abnormal situation becomes more accurate.

  Further, the movement direction calculation means and the reference movement direction calculation means may parameterize the movement direction and the reference movement direction by function approximation.

  Further, the movement direction calculation means and the reference movement direction calculation means may parameterize the movement direction and the reference movement direction using the number of Gaussian distributions and the average value and variance value of each Gaussian distribution as parameters. Good.

  According to this configuration, by expressing the parameters, it is possible to determine an abnormal situation robustly against noise and detection errors.

  The monitoring apparatus may further include a database that holds the reference movement direction set in advance.

  According to this configuration, the configuration can be simplified by setting and holding the reference movement direction in advance.

  The situation determination means compares the movement direction at the current time with the reference movement direction, and if the movement direction at the current time differs from the reference movement direction by a predetermined threshold or more, the situation at the current time is May be determined to be an abnormal situation.

  According to this configuration, it is possible to prevent a detection error, a tracking error, and a verification error by determining an abnormal situation when the moving direction of the moving object or the crowd is significantly different from the normal direction.

  In addition, the monitoring apparatus further determines that the moving direction at the current time is determined when an abnormal situation is determined by another imaging unit having at least one of pan, tilt, and zoom functions and the status determination unit. You may provide the imaging control means which controls the said other imaging means so that the area | region different from the said reference | standard movement direction may be imaged.

  According to this configuration, it is possible to obtain a detailed image by causing the other imaging unit to image the occurrence region of the abnormal situation based on the determination result of the abnormal situation.

  The imaging control unit is a region where the moving direction at the current time is different from the reference moving direction and the congestion degree has a peak value when the situation determining unit determines that an abnormal situation has occurred. The other image pickup means may be controlled so as to pick up an image.

  According to this configuration, it is possible to more accurately image the region where the abnormal situation has occurred with another imaging unit.

  Furthermore, the present invention can be realized not only as such a monitoring apparatus, but also as a monitoring method including characteristic means included in such a monitoring apparatus as steps, or causing a computer to execute these steps. It can also be realized as a program. Needless to say, such a program can be distributed via a recording medium such as a CD-ROM or a transmission medium such as the Internet.

  Embodiments of the present invention will be described below with reference to the drawings.

(Embodiment 1)
FIG. 1 is a block diagram showing the configuration of the monitoring apparatus according to Embodiment 1 of the present invention. As shown in FIG. 1, a monitoring device 100 according to Embodiment 1 of the present invention is a device that monitors a plurality of moving objects such as a moving person or vehicle or a crowd, and has a camera (imaging means) 101. A moving body image generation unit 102, a density calculation unit 103, a model generation unit 104, a database 105, a situation determination unit 106, and a notification unit 107. The notification unit 107 includes an image display unit 1071 and an alarm generation unit 1072.

  The camera 101 captures an image of an object including a plurality of moving objects or crowds moving within the monitoring area, and generates an entire image. The moving body image generation unit 102 receives the entire image captured by the camera 101 for each frame and performs inter-frame difference processing on the entire image or background difference processing with a background image prepared in advance. Here, the image subjected to the difference processing by the moving body image generation unit 102 is a moving body image in which a moving object such as a moving vehicle or a moving area of the crowd is specified. At this time, the moving body image generation unit 102 may perform the edge extraction process for each input image without performing the difference process when the edge of the image can be reliably extracted. In the following, a case where the moving body image generation unit 102 performs inter-frame difference processing or background difference processing will be described.

  The congestion degree calculation unit 103 converts the image (moving body image) that has been differentially processed by the moving body image generation unit 102 into one-dimensional information, performs frequency analysis, and shows a congestion degree that indicates the degree of congestion of the moving object or the crowd. Is calculated.

Next, the density calculation unit 103 will be described in detail with reference to FIG.
The density calculation unit 103 includes a one-dimensional information conversion unit 210 and a frequency processing unit 211. The one-dimensional information conversion unit 210 converts the two-dimensional image output from the moving body image generation unit 102 into one-dimensional information. Examples of the image conversion operation of the one-dimensional information conversion unit 210 include the following two image conversion operation examples.

  In one image conversion operation example, as shown in FIG. 3, the input image (entire image) 301 is differentially processed by the moving body image generation unit 102, and the difference value in the difference image 302 input to the density calculation unit 103 is converted into a difference value. Based on this, projection is performed on the horizontal axis, and a histogram 303 is calculated. Thereby, a histogram 303 which is one-dimensional information for each value on the horizontal axis can be obtained. In this case, one-dimensional information is calculated using the area information of the difference area.

  In another image conversion operation example, as shown in FIG. 4, a binary image 401 is generated by binarizing a difference value in an input difference image 302 to a value of “0” or “1”. One-dimensional information is obtained by scanning the digitized image 401 from the top of the image in the vertical axis direction and leaving the detected vertical axis value 402 which is the maximum value of the vertical axis of the first detected on-pixel. Here, the on-pixel is a pixel that is binarized and has a value of “1”.

  For example, when the binarized image 401 is indicated by the XY axis and the lower left is the origin, in the specific value of the X axis, there are a plurality of on-pixels on the Y axis that is the vertical axis, and the Y axis value Is “200”, “100”, “50”, the Y-axis value detected as the detection vertical axis value 402 is the Y-axis value of the on-pixel detected first after scanning from the top of the image. Therefore, “200” is obtained.

  The method for obtaining the one-dimensional information is not limited to the above, and any method may be used as long as it is a method for obtaining one-dimensional information that can calculate the crowd density.

  Here, the histogram 303 and the on-pixel vertical axis detection value 402 may be used as the crowd density. In this case, the frequency processing unit 211 is unnecessary, but in the present embodiment, as an example of further improving the robustness against noise, the frequency processing unit 211 uses the histogram 303 and the detection vertical axis as these one-dimensional information. An example in which frequency processing is performed on the value 402 will be described.

  The histogram 303 or the detected vertical axis value 402 that is one-dimensional information generated by the one-dimensional information conversion unit 210 is sent to the frequency processing unit 211. The frequency processing unit 211 performs a one-dimensional Fourier transform on the histogram 303 or the detected vertical axis value 402 that is one-dimensional information. Accordingly, the frequency spectrum 501 of the histogram 303 or the detection vertical axis value 402 that is one-dimensional information can be calculated as shown in FIG. Here, the histogram 303 which is the one-dimensional information processed by the one-dimensional information conversion unit 210 may include a high-frequency component as shown in the hatched portion of the frequency spectrum 501 shown in FIG.

  Since the monitoring device 100 is intended to capture a moving object or a crowd in a macro manner, the high frequency component is determined to be noise, and a Fourier transform is performed on the histogram 303, which is one-dimensional information, to obtain a frequency component. Is calculated, and high frequency components are cut therein, and inverse Fourier transform is performed to obtain an inverse transform result 503. In the present embodiment, the inverse transformation result 503 is referred to as one-dimensional information after frequency analysis processing. That is, the monitoring device 100 can capture a moving object or a crowd thereof macroscopically by performing low-pass processing. Here, it is not always necessary to perform the inverse Fourier transform and return to the image space, and the monitoring apparatus 100 uses the frequency spectrum 501 and the cepstrum 502 used in speech recognition, etc., so that the one-dimensional information after the frequency analysis processing is obtained. It is also possible to calculate. The cepstrum 502 is calculated by the monitoring apparatus 100 performing a one-dimensional Fourier transform, then calculating a logarithmic spectrum, performing a Fourier transform again, calculating a cepstrum coefficient, and using only the low-order component. It is also possible to calculate one-dimensional information after frequency analysis processing.

  Here, the frequency analysis-processed one-dimensional information obtained by performing frequency analysis processing such as cepstrum coefficient calculation processing on the one-dimensional information calculated by the one-dimensional information conversion unit 210 is moving. Defined as the density of moving objects or crowds.

  Of course, the histogram 303 and the on-pixel vertical axis detection value are also one-dimensional information similar to the one-dimensional information after the frequency analysis process, and can be defined as the density of a plurality of moving objects or crowds that are moving. .

  Next, the operation | movement which produces | generates the reference | standard density (model value) of the model production | generation part 104 is demonstrated. Here, the reference density is a density of a moving object or a crowd created from an entire image at normal time, and is one-dimensional information that serves as a reference for determining the occurrence of an abnormal situation.

  As a method of generating a model having a reference density, a histogram 303 and an on-pixel vertical axis detection value 402 can be used. Furthermore, in the frequency processing unit 211, there are a method using an inverse transformation result 503 obtained by low-pass processing a histogram 303 which is one-dimensional information as shown in FIG. 5, and a method using a Fourier spectrum or a cepstrum coefficient. Hereinafter, a method using an inverse transformation result 503 obtained by processing one-dimensional information with a low-pass filter will be described. Note that a method of generating a model using a Fourier spectrum or a cepstrum coefficient can also be generated using the same method as the inverse transformation result 503.

  The model generation unit 104 calculates a reference density (model value) of a moving object or a crowd based on one-dimensional information (density) after frequency analysis processing at a predetermined date and time calculated by the frequency processing unit 211. Means. The model generation unit 104 gives the calculated reference density to the database 105 to hold it.

  The model generation unit 104 can calculate the reference density for each season, time, weather, or day of the week, and can also prepare a plurality of reference densities taking into account the time of the season and the current time. .

  Further, since the number of persons or the like is considered to change depending on the weather, it is desirable to prepare a standard density according to the weather. Furthermore, since there is a person wearing an umbrella when it is outdoors in rainy weather, the one-dimensional information (density) after the frequency analysis processing is also different from that in fine weather and cloudy weather. Similarly, since it is considered that the number of persons and the like changes depending on the day of the week, such as weekdays and weekends, it is desirable to prepare a plurality of reference densities. In addition, a plurality of these standard congestions are generated so as to be associated with information on the season, time, weather or day of the week, and when monitoring, which standard congestion is used according to the current season, time, weather or day of the week. It is possible to determine whether or not.

  FIGS. 6A, 6B, and 6C are diagrams illustrating an example of modeling based on season, time, weather, and day of the week. As shown in FIGS. 6 (a), (b), and (c), for example, depending on the current season, time, weather, and day of the week, which reference density is selected from Model 1, Model 2, Model 3,. Whether to use (model value) is determined in advance. This makes it possible to model the moving object or the crowd in more detail. Note that it is not necessary to model based on the season, time, weather, and day of the week. For example, the modeling may be based on only the time. These reference densities (model values) are held in the database 105.

  Here, a reference density calculation example of the model generation unit 104 when the inverse transformation result 503 is used as the one-dimensional information (density) after frequency analysis will be described. Similar processing can be performed by using the histogram 303 and the on-pixel vertical axis detection value 402 instead of the inverse transformation result 503.

  The inverse transformation result 503 has a value y (xi) for each point xi on the x-axis. Here, the reference density (model value) is calculated by modeling the time change of y (xi), which is the inverse transformation result at each point xi. Specifically, as shown in FIG. 6 (d), by obtaining an average value and a variance value for a value that is a time change of y (xi), the time of y (xi) for each point xi on the x-axis is obtained. Variations can be modeled. That is, the reference density (model value) has an average value and a variance value of f (xi) for each point xi.

  As a result of modeling, when the variation in the density is large depending on the day of the week or the date and time, the variance value becomes a large value. In addition, when the variation in the density is small and the traffic is small, both the average value and the variance value are low, and the normal density can be modeled in consideration of the time variation. . Furthermore, it is also possible to update the variance value and the average value as the reference density sequentially.

  In addition, when a Fourier spectrum or a cepstrum coefficient is used, it is possible to model using a spectrum value at each frequency or each cepstrum order and a temporal variation of the cepstrum value.

  For example, when using a Fourier spectrum, it has a power spectrum value g (fi) for each frequency fi, so the time variation of g (fi) must be modeled using the average value and the variance value as described above. Is possible.

  Here, the database 105 includes, as the reference density (model value), an average value representing time variation of y (xi) and g (fi) of each point on the x axis or each frequency and each cepstrum order. Holds the variance value.

  Note that the model generation unit 104 calculates a reference density (model value) every predetermined time to cope with a case where the distribution of moving objects such as traffic is changed due to environmental changes, so that a new reference density ( The model density) may be obtained and given to the database 105 to update the reference density (model value).

  Next, the situation determination unit 106 will be described. The situation determination unit 106 selects a reference density (model value) held in the database 105 based on information such as the current season, time, weather, or day of the week, and calculates the selected reference density and density. A comparison is made with the density (one-dimensional information) at the current time input from the unit 103, and it is determined whether or not the density at the current time is different from the reference density, and a determination result is generated to generate a notification unit 107. To give.

  The situation determination unit 106 calculates the degree of deviation from the average value and the variance value σ of the reference density (model value) for comparison between the density at the current time and the reference density.

  Here, the occurrence of an abnormal situation is determined by generating a histogram or the like as one-dimensional information from the input image, and further using the one-dimensional information after frequency analysis processing that has been subjected to frequency analysis processing as the degree of congestion at the current time. Calculate y (xi) and g (fi) at each point on the axis or at each frequency and each cepstrum order. Next, the number of points xi in which y (xi) and g (fi) are equal to or greater than Nσ (N: constant) determined from the variance value σ of the standard density (model value) is calculated, and the number of points is equal to or greater than a threshold value. If it is, it can be judged that it is an abnormal condition. As for the comparison method, a weighting factor may be multiplied for each point xi.

  Next, the notification unit 107 will be described. The notification unit 107 includes an image display unit 1071 and an alarm generation unit 1072.

  The image display unit 1071 receives and displays the determination result from the situation determination unit 106 and notifies the monitor. The alarm generation unit 1072 receives the determination result from the situation determination unit 106 and generates an alarm sound or an alarm light when the determination result indicates the occurrence of an abnormal situation to notify the supervisor.

  Next, the operation of the monitoring apparatus 100 configured as described above will be described. FIG. 7 is a flowchart showing an operation flow of the monitoring apparatus 100. Here, it is assumed that the database 105 already holds a plurality of reference densities calculated by the model generation unit 104 according to, for example, the season, time, weather, and day of the week.

The camera 101 captures an image of an object including a plurality of moving objects or crowds moving within the monitoring area, and generates an entire image (step S101). The moving body image generation unit 102 receives the entire image captured by the camera 101 for each frame and performs inter-frame difference processing on the entire image or background difference processing with a background image prepared in advance (step S102). ). Next, the density calculation unit 103 converts the image (moving body image) that has been differentially processed by the moving body image generation unit 102 into one-dimensional information, performs frequency analysis, and calculates the density at the current time (step). S103). The situation determination unit 106 selects a reference density corresponding to the current season, time, weather, and day of the week from a plurality of reference densities held in the database 105 (step S104). Next, the situation determination unit 106 selects the selected standard density and the density calculation unit 1.
Based on the density of the current time calculated by 03, the current time situation is determined. That is, the situation determination unit 106 determines whether or not the congestion at the current time is more than a predetermined value from the average value and the variance value σ of the reference congestion (Step S105). If the result of this determination is that it exceeds a predetermined value (YES in step S105), it is determined that the current time is an abnormal situation (step S106). The alarm generation unit 1072 receives the determination result from the situation determination unit 106 as an abnormal situation, generates an alarm sound or an alarm light, and notifies the monitoring person (step S107). On the other hand, as a result of the above determination, if the predetermined value or more is not deviated (NO in step S105), the current time is determined to be the normal state (step S108).

  As described above, according to the first embodiment of the present invention, a moving object or crowd image existing in the entire image is not detected and tracked individually, but is captured in a macro manner so that a part such as a human face can be detected. Even if it cannot be detected accurately, it is possible to prevent a detection error, a tracking error, and a verification error by calling an abnormal situation when the density of moving objects or crowds is significantly different from the normal time, and alert the observer. . Further, according to Embodiment 1 of the present invention, it is easy to calculate the density of moving objects or crowds.

(Embodiment 2)
Next, Embodiment 2 of the present invention will be described with reference to the drawings. FIG. 8 is a block diagram showing the configuration of the monitoring apparatus according to Embodiment 2 of the present invention. In the second embodiment of the present invention, the same components as those in the first embodiment of the present invention are denoted by the same reference numerals, and the description thereof is omitted.

  As shown in FIG. 8, a monitoring apparatus 700 according to Embodiment 2 of the present invention includes a camera 701 and a camera control unit 702 in addition to the configuration of Embodiment 1 of the present invention.

  The camera 701 has at least one function of pan, tilt, and zoom. The density calculation unit 103 provides the camera control unit 702 with the one-dimensional information (density) after the frequency analysis processing. In addition, the situation determination unit 106 gives the determination result to the camera control unit 702. The camera control unit 702 will be described with reference to FIG. 9A shows an image example 1201 in a normal state, FIG. 9B shows an image example 1202 in the event of an abnormal situation, FIG. 9C shows an example 1203 in which the average value of the reference density is plotted, and FIG. FIG. Actually, each reference density shown in FIG. 9C has a variance value for each f (xi). Here, when the situation determination unit 106 determines that an abnormal situation has occurred, the camera control unit 702 has an area in which the one-dimensional information (density) from the density calculation unit 103 is the most different from the reference density. The camera 701 is controlled so that the camera 701 faces toward the camera. At this time, an image captured by the camera 701 is given to the image display unit 1071 and displayed.

  For example, a person falls from a normal state as shown in FIG. 9A and an abnormal situation occurs as shown in FIG. 9B, and the situation determination unit 106 determines that an abnormal situation has occurred. In this case, the camera control unit 702 moves the camera 701 so that the density 1204 at the current time as shown in FIG. 9D is most different from the reference density as shown in FIG. 9C. Control. Accordingly, the camera 701 can capture an area where the person shown in FIG.

  The camera control unit 702 can also control the camera in an area where the one-dimensional information (density) after the frequency analysis processing has a peak value.

  As described above, according to the second embodiment of the present invention, in addition to the effect of the first embodiment of the present invention, the camera 701 captures the details of the abnormal event occurrence area based on the abnormal event determination result. Can be obtained.

(Embodiment 3)
Next, Embodiment 3 of the present invention will be described with reference to the drawings. FIG. 10 is a block diagram showing the configuration of the monitoring apparatus according to Embodiment 3 of the present invention. In the third embodiment of the present invention, the same components as those in the first embodiment of the present invention are denoted by the same reference numerals, and the description thereof is omitted.

  As shown in FIG. 10, a monitoring apparatus 800 according to Embodiment 3 of the present invention adds a moving direction calculation unit 801 to the configuration of Embodiment 1 of the present invention, and generates a model generation unit 104 and a situation determination unit 106. Instead of a model generation unit 802 and a situation determination unit 803.

  The moving direction calculation unit 801 receives the image subjected to the difference processing by the moving body image generation unit 102 and performs frequency analysis processing 1 after frequency analysis processing for each frame by the frequency processing unit 211 of the density calculation unit 103. Receive dimension information.

  Hereinafter, a case will be described in which the moving direction calculation unit 801 uses, as an input, the inverse transformation result 503 obtained by performing the low-pass processing for each frame by the frequency processing unit 211 illustrated in FIG.

  As illustrated in FIG. 11, the movement direction calculation unit 801 includes an image region selection unit 901 and an image collation unit 902. As shown in FIG. 12, the image region selection unit 901 performs low-pass processing on the region of the input image 301 captured at that time based on the inverse transformation result 503 that is one-dimensional information (density) after frequency analysis processing. A region selection image 1001 is obtained by dividing the result by the inverse transformation result 503 performed. Next, the image collation unit 902 performs collation with the current input image using the region selection image 1001. At this time, the image matching unit 902 enlarges, reduces, or translates the area selection image 1001 so that the area selection image 1001 and the current input image are best matched. As a result, the image collating unit 902 obtains information on the parameters relating to enlargement or reduction and the parallel movement parameter in the vertical axis direction and the horizontal axis direction when the region selection image 1001 and the current input image are best collated. The one-dimensional array of these parameters is called the moving direction.

  Next, the image matching unit 902 gives the obtained parameter values relating to enlargement or reduction and parallel movement to the model generation unit 802 and the situation determination unit 803.

  The model generation unit 802 stores, in the database 105, one-dimensional data in which parameters relating to enlargement or reduction when the region selection image 1001 and the current input image are best matched, and parallel movement parameters in the vertical axis and horizontal axis directions are arranged. It is also possible to hold it.

  Further, as described in Embodiment 1, time variation may be taken into consideration. Specifically, the one-dimensional information obtained by arranging the parameters related to enlargement or reduction when the two area selection images 1001 are best collated and the parallel movement parameters in the vertical and horizontal axes obtained by the image matching unit 902. Further, by taking into account the time variation of each element of this one-dimensional information, the average value and the variance value of each element are calculated and held. Here, as the reference movement direction, the information held in the database 105 is one-dimensional information in which parameter values relating to enlargement, reduction, parallel movement, etc. are arranged, or one-dimensional information in which the average value and variance value of each parameter value are arranged. Information.

  Note that the image collation unit 902 may collate images by a method other than the collation method using enlargement, reduction, and parallel movement.

  The situation determination unit 803 collates the parameter value at the current time calculated by the image collation unit 902 with the parameter value (reference movement direction) calculated from the normal image held in the database 105. Here, the situation determination unit 803 selects a model (reference movement direction) held in the database 105 based on information such as the season, time, weather, or day of the week at the current time, and from the density calculation unit 103 The input information at the current time is compared.

  At this time, the situation determination unit 803 uses, as the one-dimensional information generated from the input image, the enlargement, reduction, and parallel movement used for the image matching, as in the method of generating the parameter value calculated from the normal image. The parameter value for is used. Then, the situation determination unit 803 generates an inner product value by calculating the inner product of the one-dimensional information representing the parameter value used for the image collation at the current time and the one-dimensional information calculated from the normal image, It is determined whether the inner product value is equal to or greater than the threshold value, and a determination result is generated and provided to the notification unit 107. This determination result indicates that an abnormal situation has occurred when the inner product value of the one-dimensional information representing the parameter value used for image matching and the one-dimensional information calculated from the normal image is equal to or greater than a threshold value. Is. For example, the monitoring device 800 can determine that an abnormal situation occurs when the moving object and the crowd do not move due to a conflict between the groups, even though the moving object and the crowd move at normal times.

  Here, the determination of the occurrence of the abnormal situation is made based on the inner product value of the one-dimensional information, but the similarity between the one-dimensional information calculated from the normal image and the one-dimensional information from the input image at the current time is input. It may be based on the degree. At this time, it is also effective to multiply each element by a weighting factor. In addition, the situation determination unit 803 compares the moving direction at the current time with the reference moving direction stored in the database 105, and determines that an abnormal situation occurs when the moving direction at the current time is greater than the reference moving direction by a predetermined threshold or more. Judgment.

  It is also possible to combine with Embodiment 1 of the present invention and Embodiment 3 of the present invention.

  As described above, according to the third embodiment of the present invention, a moving object or crowd image existing in the entire image is not detected and tracked individually, but is captured in a macro manner so that a part such as a human face can be detected. Even when accurate detection is not possible, it is possible to prevent a detection error, a tracking error, and a verification error by calling an abnormal situation when the moving direction of the moving object or the crowd is significantly different from the normal direction, and alert the observer. . Further, according to the third embodiment of the present invention, it is easy to calculate the moving direction of the moving object or the crowd.

(Embodiment 4)
Next, a fourth embodiment of the present invention will be described with reference to the drawings. FIG. 13 is a block diagram showing the configuration of the monitoring apparatus according to Embodiment 4 of the present invention. In the fourth embodiment of the present invention, the same components as those in the third embodiment of the present invention are denoted by the same reference numerals, and the description thereof is omitted.

  As shown in FIG. 13, a monitoring apparatus 1100 according to Embodiment 4 of the present invention includes a camera 1101 and a camera control unit 1102 in addition to the configuration of Embodiment 3 of the present invention.

  Here, an example will be described in which an image is divided into a plurality of small areas and the reference movement direction is calculated for each small area. Each process is performed for each small area. The camera 1101 has at least one function of pan, tilt, and zoom. The density calculation unit 103 provides the camera control unit 1102 with the one-dimensional information after the frequency analysis processing. In addition, the situation determination unit 803 gives the determination result to the camera control unit 1102. The camera control unit 1102 will be described with reference to FIG. FIG. 14A is a diagram showing a normal image example 1301, and shows an example in which an input image is divided into nine small regions. FIG. 14B is a diagram illustrating an example of a moving direction 1302 for each small area at the normal time. FIG. 14C is a diagram illustrating an image example 1303 when an abnormal situation occurs. FIG. 14D is a diagram illustrating an example of a moving direction 1304 when an abnormal situation occurs. Here, when the situation determination unit 106 determines that an abnormal situation has occurred, the camera control unit 1102 is calculated by the movement direction calculation unit 801 as shown in the movement direction example 1304 at the time of the abnormal situation occurrence. The camera 1101 is controlled so that the camera 1101 faces the area most different from the normal movement direction (reference movement direction) (the lower left area in the example of FIG. 14D). At this time, an image captured by the camera 1101 is given to the image display unit 1071 and displayed.

  As described above, according to the fourth embodiment of the present invention, in addition to the effects of the third embodiment of the present invention, the camera 1101 captures the details of the abnormal event occurrence area based on the abnormal event determination result. Can be obtained.

  In this embodiment, the camera control unit 1102 controls the camera 1101 so that the movement direction calculated by the movement direction calculation unit 801 is directed to the area where the movement direction is the most different from the reference movement direction. It is not limited to. For example, the camera control unit 1102 may change the camera 1101 so that the camera 1101 faces a region where the movement direction calculated by the movement direction calculation unit 801 is different from the reference movement direction by a predetermined threshold or more and where one-dimensional information has a peak value. 1101 may be controlled. Also, the camera control unit 1102 may control the camera 1101 so that the camera 1101 is sequentially directed to a plurality of areas in which the movement direction calculated by the movement direction calculation unit 801 is different from the reference movement direction by a predetermined value or more, for example. .

(Embodiment 5)
In the present embodiment, a mode in which the density is calculated from the input image using the area of the human head candidate region and the area occupied by the person will be described. The configuration of the monitoring apparatus according to Embodiment 5 of the present invention is the same as that shown in FIG.

  The camera 101 captures an image of an object including a plurality of moving objects or crowds that are moving, and generates an entire image. Here, it is desirable to use a color image or an infrared image. The moving body image generation unit 102 may perform background difference processing.

  Next, a case where the density calculation unit 103 performs head detection will be described. For example, a black area (or a skin color area) is extracted from an entire image 1601 as shown in FIG. 15A photographed by the camera 101 to obtain a detection result 1602 as shown in FIG. 15B, for example. Next, by detecting an ellipse corresponding to the head, for example, an ellipse detection result 1603 as shown in FIG. 15C is obtained, and the area of the head candidate region included in the image is calculated. Here, in order to calculate the density in a macro manner, even when the detection of the ellipse corresponding to the head does not operate normally, such as when people overlap, as in the example shown in FIG. The operation does not fail. Even when the number of heads cannot be detected accurately, it can be expected to operate normally.

  Next, a case where the density calculation unit 103 calculates the area occupied in the human image will be described. The camera 101 preferably uses an infrared image. In this case, the density calculation unit 103 calculates the number of pixels occupied by the person area detected in the infrared image. Of course, the person area may be cut out after skin color detection or the like is performed on the color image. Here, as in the case of detecting the area of the head candidate region, the person region does not necessarily have to be extracted accurately.

  Here, the degree of congestion can be defined by the area of the head candidate region or the number of pixels occupied by the person region in the image, the ratio, and the like.

  Next, the operation | movement which produces | generates the reference | standard density (model value) of the model production | generation part 104 is demonstrated. Here, the reference density is a density of a moving object or a crowd created from an entire image at normal time, and an area of a head candidate area or a person area that is a reference for determining occurrence of an abnormal situation is included in the image. This is the number and ratio of pixels.

  In order to cope with temporary fluctuations in the density, an average value and a variance value for a certain time may be used as the reference density. The model generation unit 104 gives the calculated reference density to the database 105 to hold it. Further, since the number of persons or the like is considered to change depending on the weather, it is desirable to prepare a standard density according to the weather. Furthermore, since there is a person wearing an umbrella when it is outdoors in rainy weather, the one-dimensional information (density) after the frequency analysis processing is also different from that in fine weather and cloudy weather. Similarly, since it is considered that the number of persons and the like changes depending on the day of the week, such as weekdays and weekends, it is desirable to prepare a plurality of reference densities. In addition, a plurality of these standard congestions are generated so as to be associated with information on the season, time, weather or day of the week, and when monitoring, which standard congestion is used according to the current season, time, weather or day of the week. It is possible to determine whether or not.

  Next, the situation determination unit 106 will be described. The situation determination unit 106 selects a reference density (model value) held in the database 105 based on information such as the current season, time, weather, or day of the week, and calculates the selected reference density and density. A comparison is made with the density (one-dimensional information) at the current time input from the unit 103, and it is determined whether or not the density at the current time is different from the reference density, and a determination result is generated to generate a notification unit 107. To give.

  The situation determination unit 106 calculates the degree of deviation from the average value and the variance value σ of the reference density (model value) for comparison between the density at the current time and the reference density. As an example, an abnormal situation can be determined when the value of the density is more than Nσ (N: constant) determined from the variance value σ of the standard density (model value).

  Next, the notification unit 107 will be described. The notification unit 107 includes an image display unit 1071 and an alarm generation unit 1072.

  The image display unit 1071 receives and displays the determination result from the situation determination unit 106 and notifies the monitor. The alarm generation unit 1072 receives the determination result from the situation determination unit 106 and generates an alarm sound or an alarm light when the determination result indicates the occurrence of an abnormal situation to notify the supervisor.

  As described above, according to the fifth embodiment of the present invention, the density of moving objects or crowds is captured by capturing macroscopic images of moving objects or crowds existing in the entire image instead of individually detecting and tracking them. By detecting that the situation is abnormal when it is significantly different from the normal time, it is possible to prevent a detection error, a tracking error, and a verification error, and call attention to the monitor. Further, according to the fifth embodiment of the present invention, the amount of calculation is small compared to the first embodiment, and it becomes easy to calculate the density of moving objects or crowds.

(Embodiment 6)
Next, a sixth embodiment of the present invention will be described. The configuration of the monitoring apparatus according to Embodiment 6 of the present invention is the same as that shown in FIG.

  In the present embodiment, the camera (imaging means) 101, the moving body image generation unit 102, and the density calculation unit 103 are the same as those in the first embodiment, and thus description thereof is omitted.

  The model generation unit 104 can simply calculate using the arithmetic mean and variance when generating the reference density (model value), but here, the reference density is calculated using function approximation. An example of application will be described.

  This background image creation method is a method of learning a sequentially input image by probabilistically modeling a background image using a plurality of Gaussian distributions for each pixel. For example, there is an advantage that pixels that are constantly changing, such as shaking of trees, can be used as a background including the change. In the model generation unit 104 in the monitoring apparatus 100 of the present invention, a model that can cope with environmental changes can be generated by considering the one-dimensional information that is normally input as reference information in the same way as the background. it can.

  Here, the reference density is a density of a moving object or a crowd created from an entire image at normal time, and is one-dimensional information that serves as a reference for determining the occurrence of an abnormal situation.

  As a method of generating a model having a reference density, a histogram 303 and an on-pixel vertical axis detection value 402 can be used. Furthermore, in the frequency processing unit 211, there are a method using an inverse transformation result 503 obtained by low-pass processing a histogram 303 which is one-dimensional information as shown in FIG. 5, and a method using a Fourier spectrum or a cepstrum coefficient. Hereinafter, a method using an inverse transformation result 503 obtained by processing one-dimensional information with a low-pass filter will be described. Note that a method of generating a model using a Fourier spectrum or a cepstrum coefficient can also be generated using the same method as the inverse transformation result 503.

  The model generation unit 104 calculates and calculates a reference density (model value) of a moving object or crowd based on one-dimensional information (density) after frequency analysis processing at a predetermined date and time calculated by the frequency processing unit 211. It constitutes a calculation means. The model generation unit 104 gives the calculated reference density to the database 105 to hold it.

  The model generation unit 104 can calculate the reference density for each season, time, weather, or day of the week, and can also prepare a plurality of reference densities taking into account the time of the season and the current time. .

  Further, since the number of persons or the like is considered to change depending on the weather, it is desirable to prepare a standard density according to the weather. Furthermore, since there is a person wearing an umbrella when it is outdoors in rainy weather, the one-dimensional information (density) after the frequency analysis processing is also different from that in fine weather and cloudy weather. Similarly, since it is considered that the number of persons and the like changes depending on the day of the week, such as weekdays and weekends, it is desirable to prepare a plurality of reference densities. In addition, a plurality of these standard congestions are generated so as to be associated with information on the season, time, weather or day of the week, and when monitoring, which standard congestion is used according to the current season, time, weather or day of the week. It is possible to determine whether or not.

  Here, a reference density calculation example of the model generation unit 104 when the inverse transformation result 503 is used as the one-dimensional information (density) after frequency analysis will be described. Similar processing can be performed by using the histogram 303 and the on-pixel vertical axis detection value 402 instead of the inverse transformation result 503.

  The inverse transformation result 503 has a value y (xi) for each point xi on the x-axis. Here, the reference density (model value) is calculated by modeling the time change of y (xi), which is the inverse transformation result at each point xi. Here, C.I. Stauffer, W.M. E. L. Grimsson: “Adaptive background mixture models for real-time tracking,” CVPR '99, Vol. 2, pp. An example using the background image creation method in 246-252 will be described.

  Here, for a time change of y (xi), fitting is performed using, for example, a plurality of Gaussian distributions as indicated by dotted lines in FIG. Thereby, the time change of y (xi) can be expressed as a parameter indicating the number of Gaussian distributions and the average value and variance of each Gaussian distribution.

  In this embodiment, this parameter set is referred to as a reference density (model value) and is stored in the database 105.

  As a result of modeling, when the variation in the density is large depending on the day of the week or the date and time, the variance value becomes a large value. In addition, when the variation in the density is small and the traffic is small, both the average value and the variance value are low, and the normal density can be modeled in consideration of the time variation. . Further, by modeling with a plurality of Gaussian distributions, it is also possible to model the density that periodically fluctuates due to changes in traffic lights when monitoring the vicinity of traffic lights.

  In addition, when a Fourier spectrum or a cepstrum coefficient is used, it is possible to model using a spectrum value at each frequency or each cepstrum order and a temporal variation of the cepstrum value.

  For example, when a Fourier spectrum is used, since it has a power spectrum value g (fi) for each frequency fi, the time variation of g (fi) can be represented by the number of Gaussian distributions and the average and variance values of each Gaussian distribution. Can be expressed as a set.

  Here, the parameter set is stored in the database 105 as a reference density (model value). Of course, the parameter set can be updated every time.

  Next, the situation determination unit 106 will be described. The situation determination unit 106 selects a reference density (model value) held in the database 105 based on information such as the current season, time, weather, or day of the week, and calculates the selected reference density and density. A comparison is made with the density (one-dimensional information) at the current time input from the unit 103, and it is determined whether or not the density at the current time is different from the reference density, and a determination result is generated to generate a notification unit 107. To give.

  The situation determination unit 106 calculates the degree of deviation from the average value and the variance value σ of the reference density (model value) for comparison between the density at the current time and the reference density.

  Here, the occurrence of an abnormal situation is determined by generating a histogram or the like which is one-dimensional information from the input image, and further, as a degree of congestion at the current time from the one-dimensional information after frequency analysis processing that has been subjected to frequency analysis processing, x Calculate y (xi) and g (fi) at each point on the axis or at each frequency and each cepstrum order. Next, whether y (xi) and g (fi) are abnormal is determined by setting a threshold value and a range thereof from a reference density (model value) modeled by a Gaussian distribution.

  Next, the notification unit 107 will be described. The notification unit 107 includes an image display unit 1071 and an alarm generation unit 1072.

  The image display unit 1071 receives and displays the determination result from the situation determination unit 106 and notifies the monitor. The alarm generation unit 1072 receives the determination result from the situation determination unit 106 and generates an alarm sound or an alarm light when the determination result indicates the occurrence of an abnormal situation to notify the supervisor.

  Note that the model generation unit 104 calculates a reference density (model value) every predetermined time to cope with a case where the distribution of moving objects such as traffic is changed due to environmental changes, so that a new reference density ( The model density) may be obtained and given to the database 105 to update the reference density (model value).

  As described above, according to the sixth embodiment of the present invention, a moving object or crowd image existing in the entire image is not detected and tracked individually, but is captured in a macro manner, so that a part such as a human face can be identified. Even if it cannot be detected accurately, it is possible to prevent a detection error, a tracking error, and a verification error by calling an abnormal situation when the density of moving objects or crowds is significantly different from the normal time, and alert the observer. . Furthermore, according to the sixth embodiment of the present invention, it becomes easy to calculate the density of moving objects or crowds, and in addition to the effects of the first embodiment, it is possible to more accurately model time fluctuations.

  The monitoring device according to the present invention is useful for determining an abnormal situation of a moving object or a crowd in, for example, a station, a shopping street, a road, or the like.

It is a block diagram which shows the structure of the monitoring apparatus which concerns on Embodiment 1 of this invention. It is a block diagram which shows the structure of the density calculation part of the monitoring apparatus which concerns on Embodiment 1 of this invention. It is a figure for demonstrating the one-dimensional information conversion in the monitoring apparatus which concerns on Embodiment 1 of this invention. It is another figure for demonstrating the one-dimensional information conversion in the monitoring apparatus which concerns on Embodiment 1 of this invention. It is a figure for demonstrating the low-pass process in the monitoring apparatus which concerns on Embodiment 1 of this invention. It is a figure for demonstrating the production | generation of the model in the monitoring apparatus which concerns on Embodiment 1 of this invention. It is a flowchart which shows the flow of operation | movement of the monitoring apparatus which concerns on Embodiment 1 of this invention. It is a block diagram which shows the structure of the monitoring apparatus which concerns on Embodiment 2 of this invention. It is a figure for demonstrating operation | movement of the camera control part of the monitoring apparatus which concerns on Embodiment 2 of this invention. It is a block diagram which shows the structure of the monitoring apparatus which concerns on Embodiment 3 of this invention. It is a block diagram which shows the structure of the moving direction calculation part of the monitoring apparatus which concerns on Embodiment 3 of this invention. It is a figure for demonstrating operation | movement of the monitoring apparatus which concerns on Embodiment 3 of this invention. It is a block diagram which shows the structure of the monitoring apparatus which concerns on Embodiment 4 of this invention. It is a figure for demonstrating operation | movement of the camera control part of the monitoring apparatus which concerns on Embodiment 4 of this invention. It is a figure for demonstrating operation | movement of the head candidate area | region detection in the monitoring apparatus which concerns on Embodiment 5 of this invention. It is a figure for demonstrating modeling by the Gaussian distribution in the monitoring apparatus which concerns on Embodiment 6 of this invention. It is a block diagram which shows the structure of the conventional monitoring apparatus.

Explanation of symbols

100, 700, 800, 1100 Monitoring device 101, 701, 1101 Camera 102 Moving body image generation unit 103 Concentration calculation unit 104, 802 Model generation unit 105 Database 106, 803 Situation determination unit 107 Notification unit 210 One-dimensional information conversion unit 211 Frequency processing unit 702, 1102 Camera control unit 801 Movement direction calculation unit 901 Image area selection unit 902 Image collation unit 1071 Image display unit 1072 Alarm generation unit

Claims (10)

A monitoring device that performs monitoring in a monitoring area,
Imaging means for capturing an image in the monitoring area and generating an entire image;
Moving body image generating means for generating a moving body image that is a plurality of moving objects or crowd images from the whole image generated by the imaging means;
A density calculation means for calculating a density indicating the degree of congestion of the moving object or the crowd based on the moving body image generated by the moving body image generation means;
The density at the current time calculated by the density calculation means is compared with a standard density that is a density as a criterion for situation determination, and the density at the current time and the reference density are determined to be predetermined. A situation judging means for judging that the situation at the current time is an abnormal situation when different from the threshold, and
Other imaging means having at least one of pan, tilt and zoom functions;
An imaging control means for controlling the other imaging means so as to capture an area where the density at the current time is different from the reference density when the situation determination means determines that the situation is abnormal. A monitoring device characterized by. The monitoring device further includes:
The monitoring apparatus according to claim 1, further comprising a notification unit that notifies the status at the current time determined by the status determination unit. The monitoring apparatus according to claim 1, wherein the moving body image generation unit generates the moving body image by performing inter-frame difference processing or background difference processing on the entire image. The density calculation unit detects a head candidate area from the whole image, and calculates the density based on an area occupied by the detected head candidate area with respect to the whole image. Item 1. The monitoring device according to Item 1. The monitoring apparatus according to claim 1, wherein the density calculation unit calculates the density based on an area of the moving body image. The monitoring apparatus according to claim 1, wherein the density calculation unit converts the moving body image into one-dimensional information, and calculates the density based on the converted one-dimensional information. The monitoring device according to claim 6, wherein the density calculation unit calculates the density by performing a frequency analysis of the one-dimensional information. The imaging means is an infrared camera;
The monitoring apparatus according to claim 1, wherein the density calculation unit calculates the density based on a reaction area of the infrared camera. A monitoring method for monitoring in a monitoring area,
An imaging step of capturing an image in the monitoring area by an imaging means to generate an entire image;
A moving body image generating step for generating a moving body image that is an image of a plurality of moving objects or a crowd from the entire image generated by the imaging step;
A density calculation step of calculating a density indicating a degree of congestion of the moving object or the crowd based on the moving object image generated by the moving object image generation step;
The density at the current time calculated by the step of calculating the density is compared with a standard density that is a density as a criterion for situation determination, and the density at the current time and the reference density are determined to be predetermined. A situation determination step for determining that the situation at the current time is an abnormal situation when different from the threshold, and
When the situation determination step determines that an abnormal situation has occurred, another imaging unit having at least one of a pan, tilt, and zoom function captures an area where the density at the current time is different from the reference density. As described above, a monitoring method comprising: an imaging control step of controlling the other imaging means. A program for monitoring in a monitoring area,
An imaging step of capturing an image in the monitoring area by an imaging means to generate an entire image;
A moving body image generating step for generating a moving body image that is an image of a plurality of moving objects or a crowd from the entire image generated by the imaging step;
A density calculation step of calculating a density indicating a degree of congestion of the moving object or the crowd based on the moving object image generated by the moving object image generation step;
The density at the current time calculated by the step of calculating the density is compared with a standard density that is a density as a criterion for situation determination, and the density at the current time and the reference density are determined to be predetermined. A situation determination step for determining that the situation at the current time is an abnormal situation when different from the threshold, and
When the situation determination step determines that an abnormal situation has occurred, another imaging unit having at least one of a pan, tilt, and zoom function captures an area where the density at the current time is different from the reference density. As described above, a program for causing a computer to execute an imaging control step of controlling the other imaging means.

Images