JP2006099241A - Abnormality detection apparatus, abnormality detection method and abnormality detection program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To detect local changes in the state of crowds even in cases where the flow of people is complex in terms of its direction and state and varies greatly with time. <P>SOLUTION: An optical flow calculating part 2 calculates the optical flow (movement vector) of the current frame on the basis of image data 3. A flow time average calculating part 5 sets the optical flow at its absolute value and calculates the time average of the optical flow to obtain pixel flow for each pixel (average in time direction). A block average calculating part 6 averages the pixel flow within each designated block to obtain average flow (average in space direction). An average flow comparing part 10 compares the average flows between the blocks and determines whether or not a difference between the average flows is great according to a threshold designated by a threshold designating part 9. An abnormality detecting part 11 detects a great difference between the average flows as an abnormality. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an abnormality detection device, an abnormality detection method, and an abnormality detection program for detecting a change in the state of a crowd when some trouble occurs in a place where people gather, and detecting this by image processing.

  In a place where many people gather, various troubles occur, and it is necessary to detect these troubles from the viewpoints of safety management and disaster prevention, and take prompt measures. Therefore, there is an expectation for a technology that reduces the burden on the observer by automatically detecting troubles using the video of the fixed point camera.

As a conventional technique, there is a technique that uses a motion vector (optical flow) based on a fixed-point camera image (see, for example, Patent Document 1). In this conventional technique, the average (average vector) and variance of the motion vector in the normal area in the attention area (block) are obtained, and an abnormality is detected when the difference from the average vector calculated at the time of observation increases. To do.
JP 2001-357484 A

  The above-mentioned conventional technique is effective when the flow is fixed (like the direction of the flow is fixed) like a car on a road, but it is the direction like a crowd at a station concourse. There is a problem that it is difficult to apply when the number of people, the speed, etc. change variously with time.

  The present invention has been made in consideration of such circumstances, and the purpose of the present invention is to make the direction and state complex and change variously over time, such as the flow of people in a station concourse. Even in such a case, an object of the present invention is to provide an abnormality detection device, an abnormality detection method, and an abnormality detection program that can detect that the state of the crowd has changed locally.

  The present invention has been made to solve the above-described problems. An input means for inputting a moving image to be monitored for each frame image and an optical flow in each image are obtained in a target area set in advance in the frame image. A pixel flow calculation unit that calculates a pixel flow indicating the movement of each pixel from the optical flow, and a state index that indicates a feature of the spatial distribution of the pixel flow in the frame image is calculated from the calculated pixel flow. An abnormality detection apparatus comprising: a state index calculation unit; and an abnormality detection unit that detects an abnormality of the monitoring target based on the state index.

  Further, in the present invention, the present invention provides at least two or more target areas in one frame image, and the state index calculation means calculates the average flow of each target area based on each pixel flow in the target area. And the state index is calculated based on the difference in the average flow of each target region.

  According to the present invention, in the abnormality detection device according to claim 2, the state index calculation unit calculates the state index based on a variance of each pixel flow in the target region.

  Further, the present invention is characterized in that, in the above-mentioned invention, the state index calculating means calculates the state index based on a spatial change in each pixel flow in the target region.

  Further, the present invention is characterized in that, in the above-mentioned invention, the pixel flow calculation means calculates an average of the optical flow of each pixel in a past fixed time from the current frame image as each pixel flow.

  Further, the present invention is characterized in that, in the above-mentioned invention, the pixel flow calculation means calculates each pixel flow based on an absolute value of an optical flow of each pixel.

  Further, the present invention inputs a moving image to be monitored for each frame image, obtains an optical flow in each image in a target region set in advance in the frame image, and indicates the movement of each pixel from the optical flow. Calculating a pixel flow, calculating a state index indicating a feature of a spatial distribution of the pixel flow in the frame image from the calculated pixel flow, and detecting an abnormality of the monitoring target based on the state index Is an abnormality detection method characterized by

  The present invention also includes a step of inputting a moving image to be monitored for each frame image, an optical flow in each image in a target region preset in the frame image, and a motion of each pixel from the optical flow. A step of calculating a pixel flow indicating the state, a step of calculating a state index indicating a feature of a spatial distribution of the pixel flow in the frame image from the calculated pixel flow, and the monitoring target based on the state index An abnormality detection program for causing a computer to execute a step of detecting an abnormality in the computer.

  As described above, the anomaly detection apparatus inputs a moving image to be monitored for each frame image, obtains an optical flow in each image in a target region set in advance in the frame image, and from each optical flow, A pixel flow indicating the movement of the pixel is calculated, and from the calculated pixel flow, a state index indicating the characteristics of the spatial distribution of the pixel flow in the frame image is calculated, and an abnormality of the monitoring target is calculated based on the state index. Detect. Therefore, it is possible to detect an abnormality of the monitoring target from the state index indicating the feature of the spatial distribution of the pixel flow calculated from the optical flow in the target region. As a result, it is possible to detect that the state of the crowd has changed locally even when the direction and state of the station is complicated and changes with time, such as the flow of people in a concourse at a station. The effect of being able to be obtained.

  According to the present invention, the abnormality detection device provides at least two or more target areas in one frame image, calculates an average flow of each target area based on each pixel flow in the target area, A state index is calculated based on the difference in the average flow of the area. For this reason, it is possible to compare stable state indexes with the average flow obtained by averaging the pixel flow in the time direction and the spatial direction, and to easily detect an abnormality of the monitoring target.

  Further, according to the present invention, the abnormality detection device calculates the state index based on the variance of each pixel flow in the target region. For this reason, it is possible to evaluate the variation in pixel flow in the target area by statistical means and detect an abnormality of the monitoring target.

  According to the present invention, the abnormality detection device calculates a state index based on a spatial change in each pixel flow in the target region. Therefore, it is possible to calculate a state index using a means such as a high-pass filter for a spatial change of the pixel flow in the target region, that is, a spatial inclination, and to detect an abnormality of the monitoring target. .

  Further, according to the present invention, the abnormality detection device calculates the average of the optical flow of each pixel in the past fixed time from the current frame image as each pixel flow. Therefore, the effect that the state index can be stabilized by averaging is obtained.

  According to the invention, the abnormality detection device calculates each pixel flow based on the absolute value of the optical flow of each pixel. Therefore, it is possible to easily detect an abnormality even in a place where flows in various directions are mixed.

  Hereinafter, the best mode for carrying out the present invention will be described. In the following description, the optical flow is a so-called motion vector obtained for each pixel in the target region, and can be obtained from the current frame image and the previous frame image immediately before the current frame image. The pixel flow is a value indicating the movement of each pixel obtained from the optical flow. In the first embodiment described later, the pixel flow is obtained for each pixel by averaging the absolute value of each optical flow for a certain past time. The average flow is obtained by averaging the image flows in the blocks corresponding to the above-described target region, and is obtained for each block. The state index is an index for determining whether or not there is an abnormality. In the first embodiment, which will be described later, the state index is obtained from the difference in average flow between blocks. In the second embodiment, which will be described later, In the third embodiment, which will be described later, the pixel flow distribution is obtained by the spatial change of the pixel flow in the target region.

A. First embodiment A-1. Configuration of First Embodiment FIG. 1 is a block diagram showing a schematic configuration of an abnormality detection apparatus according to a first embodiment of the present invention. In FIG. 1, the pixel flow calculation unit 100 a includes a block designation unit 1, an optical flow calculation unit 2, a flow recording unit 4, and a flow time average calculation unit 5. The state index calculation unit 101a includes a flow block average calculation unit 6, a threshold designation unit 9, and an average flow comparison unit 10. In the pixel flow calculation unit 100a, the block specifying unit 1 specifies a plurality of processing target blocks to be processed in the frame. The optical flow calculation unit 2 calculates an optical flow (motion vector) of the current frame from the video data 3 including a plurality of frame image sequences. The flow recording unit 4 records an optical flow for a predetermined time calculated by the optical flow calculation unit 2. The flow time average calculation unit 5 converts the optical flow into absolute values and calculates the time average of the optical flow.

  In the state index calculation unit 101a, the flow block average calculation unit 6 averages the time averaged flow for each pixel, that is, the pixel flow within each designated block. In the figure, averaging is performed for block A and block B as designated blocks, and an average flow 7 of block A and an average flow 8 of block B are output. The threshold designation unit 9 designates a threshold when comparing the average flows of designated blocks described later. The average flow comparison unit 10 compares the average flows 7 and 8 between the blocks, and determines whether or not the difference between the average flows 7 and 8 is large according to the threshold specified by the threshold specification unit 9. The abnormality detection unit 11 detects an abnormality when the difference in the average flow is large.

A-2. Operation of First Embodiment Next, the operation of the abnormality detection device according to the first embodiment described above will be described. Here, FIG. 2 is a flowchart for explaining the operation of the abnormality detection device according to the first embodiment.
First, a plurality of blocks to be processed are specified by the block specifying unit 1 (Sa1). Next, among the video data composed of a plurality of frame image series, the target frame is advanced by one (Sa2). Next, the optical flow calculation unit 2 calculates the optical flow of the current frame (Sa3). The calculated optical flow is recorded in the flow recording unit 4. The optical flow is calculated as optical flow = (u, v) ⁿ x, y at the coordinates (x, y) of the frame n, where the coordinates on the image are (x, y). Note that u is the x component of the velocity vector, and v is the y component of the velocity vector.

Next, the flow time average calculation unit 5 reads the optical flow from the flow recording unit 4, calculates the absolute value of the optical flow for each x and y component (Sa4), and calculates the time average of the optical flow in the past fixed time. (Sa5). The absolute value of the optical flow is (| u |, | v |) ⁿ x, y. Further, the time average (past N frames) of the optical flow is (U, V) x, y = Σ (| u |, | v |) ^kx , y / N. Note that k = n−N to n. In this way, the pixel flow for each pixel is calculated from the optical flow.
In general, the absolute value of the vector (a, b) often indicates √ (a ² + b ² ) which is the length of the vector, but in this embodiment, the absolute value (| a Either |, | b |) or the vector length √ (a ² + b ² ) may be used. In the following description, the case of absolute values for each component (| a |, | b |) will be described as an example.

  Next, the block average calculation unit 6 averages the pixel flows within each designated block to obtain an average flow 7 of block A and an average flow 8 of block B (Sa6). The average (average flow) within the block of pixel flow is A￣ = (Ua, Va) = Σ (U, V) x, y / Na for block A (Na is the number of pixels in block A, x and y are coordinates in block A), and for block B, B￣ = (Ub, Vb) = Σ (U, V) x, y / Nb (Nb is the number of pixels in block B, x , Y are coordinates in block B). The order of the time average and the block average may be reversed. “￣” indicates a vector.

  The average flow here is the result of averaging the absolute value of the calculated optical flow in the time direction and the spatial direction in each block, and one average flow (vector) is calculated for each block. FIG. 3A is an original image, and FIG. 3B is a diagram showing that two blocks are designated and an average flow is obtained for each block.

  Next, the average flow comparison unit 10 compares the average flow 7 of the block A with the average flow 8 of the block B (Sa7), and according to the threshold value Ot specified by the threshold value specification unit 9, the difference between both average flows is determined. It is determined whether it is large (Sa8), and the evaluation result is transmitted to the abnormality detection unit 11. As an example of a method for evaluating the difference in average flow, O = A￣ · B￣ / max [A￣ · A￣, B￣ · B￣] (• is an inner product), and the comparison result O is a preset threshold value. When it is smaller than Ot, it is determined that the difference is large. “￣” indicates a vector.

  The number of blocks to be compared can be expanded to 3 or more as needed, and adjacent locations and locations where the correlation of the human state is high, such as the same space characteristics (such as on the same passage), are selected. It is preferable.

  Next, when the difference is large according to the evaluation result, the abnormality detection unit 11 detects that the abnormality is detected (Sa9), determines whether the end condition is satisfied (Sa10), and if not, Returning to Sa2, the target frame is advanced to the next frame, and the above-described processing is repeated.

  On the other hand, if the difference between the two is small, it is determined whether or not the end condition is satisfied without detecting an abnormality (Sa10). If not, the process returns to step Sa2 to advance the target frame to the next frame. The above process is repeated. Further, when the termination condition is satisfied, the process is terminated.

  In addition, this method can be applied even when the average flow varies depending on the location at normal times (eg, when the density of people in block A is about 40% of the density of people in block B). . In such a case, a normal average flow may be calculated in advance and normalized so that the size of the average flow is equal between blocks. In addition, it is possible to stabilize the comparison between perspectives by correcting in advance the difference in appearance in the image, for example, the smaller the distance, the smaller the movement on the image, the slower the movement on the image. .

B. Second Embodiment Next, a second embodiment of the present invention will be described.
B-1. Configuration of Second Embodiment FIG. 4 is a block diagram showing a configuration of an abnormality detection apparatus according to the second embodiment. In the figure, the pixel flow calculation unit 100 b includes a processing target area designating unit 20, an optical flow calculation unit 21, a flow recording unit 23, and a flow time average calculation unit 24. The state index calculation unit 101 b includes a flow blur processing unit 25, a flow statistic calculation unit 26, a threshold specification unit 27, and a threshold comparison unit 28. In the pixel flow calculation unit 100b, the processing target area designating unit 20 designates the target area to be monitored, that is, the processing target area. The optical flow calculation unit 21 calculates the optical flow of the current frame from the video data 22 including a plurality of frame image series. The flow recording unit 23 records the optical flow calculated by the optical flow calculation unit 21. The flow time average calculation unit 24 converts the optical flow into absolute values and calculates the time average of the optical flow to obtain a pixel flow.

  In the state index calculation unit 101b, the flow blur processing unit 25 performs a process of blurring in the spatial direction. A Gaussian filter or the like is used for the blurring process. The flow statistic calculation unit 26 calculates the variation (dispersion) of the pixel flow in the processing target area subjected to the blurring process. As described above, in the second embodiment, as a method for evaluating the variation in pixel flow in the target region, a method using basic statistics such as average, variance, or standard deviation (evaluation by statistical processing) is employed. ing. The threshold value designation unit 27 designates a threshold value to be compared with pixel flow variations described later. The threshold comparison unit 28 obtains the variation in pixel flow within the processing target region, and determines whether the variation in pixel flow is large according to the threshold specified by the threshold specification unit 27. The abnormality detection unit 29 detects an abnormality when the variation in pixel flow is large.

B-2. Operation of Second Embodiment Next, the operation of the abnormality detection device according to the second embodiment described above will be described. Here, FIG. 5 is a flowchart for explaining the operation of the abnormality detection device according to the second embodiment. FIG. 6 is a conceptual diagram for explaining the operation of the abnormality detection device according to the second embodiment.

First, among the video data composed of a plurality of frame image series, the target frame is advanced by one (Sb1). Next, the optical flow calculation unit 21 calculates the optical flow of the current frame according to the processing target area specified by the processing target area specifying unit 20 (Sb2). The calculated optical flow is recorded in the flow recording unit 23. The optical flow is calculated as optical flow = (u, v) ⁿ x, y at the coordinates (x, y) of the frame n, where the coordinates on the image are (x, y). Note that u is the x component of the velocity vector, and v is the y component of the velocity vector.

Next, the flow time average calculation unit 24 reads the optical flow from the flow recording unit 23, calculates the absolute value of the optical flow for each of the x and y components (Sb3), and calculates the time average of the optical flow in a past fixed time. (Sb4). The absolute value of the optical flow is (| u |, | v |) ⁿ x, y. Further, the time average (past N frames) of the optical flow is (U, V) x, y = Σ (| u |, | v |) ^k x, y. Note that k = n−N to n). In this way, a pixel flow for each pixel is obtained from the optical flow.

  Next, the blurring processing unit 25 blurs the pixel flow in the spatial direction (Sb5). The pixel flow here is the result of performing the time direction average and blurring processing (equivalent to the weighted average of neighboring pixels) on the absolute value of the calculated optical flow, and each pixel in the processing target area The pixel flow (vector) is calculated at (see the arrow in FIG. 6A).

Next, the flow statistic calculation unit 26 calculates the variation (dispersion) of the pixel flow in the processing target area subjected to the blurring process (Sb6). Here, FIG. 6B shows the appearance frequency in the processing target region with respect to the absolute value | u | of the pixel flow x component in each pixel. Note that the absolute value | u | of the pixel flow x component is the absolute value √ (u ² + v ² ) of the pixel flow at each pixel even if it is the absolute value | v | of the pixel flow y component at each pixel. There may be. However, u and v are pixel flow values after being blurred (smoothed or averaged) in the spatial direction and the temporal direction.

  Next, the threshold value comparison unit 28 obtains the variation in pixel flow between the processing target regions, determines whether the variation in pixel flow is large according to the threshold value designated by the threshold value designation unit 27 (Sb7), and the evaluation result Is transmitted to the abnormality detection unit 29. Next, when the variation in pixel flow is large according to the evaluation result, the abnormality detection unit 29 detects that the abnormality is present (Sb8), determines whether or not the termination condition is satisfied (Sb9), and ends the process. For example, the process returns to step Sb1, the target frame is advanced to the next frame, and the above-described processing is repeated.

  On the other hand, if the difference between the two is small, it is determined whether or not the end condition is satisfied without detecting an abnormality (Sb9). If not, the process returns to step Sb1 to advance the target frame to the next frame. The above process is repeated. Further, when the termination condition is satisfied, the process is terminated.

  In addition, if the difference in appearance in the image (the farther it looks, the smaller the movement, and the farther the movement on the image looks slower) is corrected in advance, it is possible to stabilize the processing results for the range including both near and far. . In addition, this method is expanded when the pixel flow that occurs in the processing target area in normal conditions varies depending on the location (for example, the upper left person density is about 40% of the lower right person density). can do. In such a case, the normal pixel flow may be calculated in advance, and the size of the pixel flow may be normalized within the processing target area.

C. Third Embodiment Next, a third embodiment of the present invention will be described. The abnormality detection device according to the third embodiment is different from the configuration shown in FIG. 4 except that the flow statistic calculation unit 26 detects a spatial inclination. As described above, in the third embodiment, as a technique for evaluating the variation of the pixel flow in the target region to be monitored, that is, the processing target region, a differential filter (high-pass filter) process such as a Sobel filter is performed, A method of detecting a rapid change (amount of change or rate of change) in the spatial direction of the magnitude of the pixel flow within the target region is employed.

C-1. Operation of Third Embodiment Next, the operation of the abnormality detection device according to the third embodiment described above will be described. Here, FIG. 7 is a flowchart for explaining the operation of the abnormality detection apparatus according to the third embodiment. FIG. 8 is a conceptual diagram for explaining the operation of the abnormality detection device according to the third embodiment.

  First, among the video data composed of a plurality of frame image series, the target frame is advanced by one (Sc1). Next, the optical flow calculation unit 21 calculates the optical flow of the current frame according to the processing target area specified by the processing target area specifying unit 20 (Sc2).

  Next, the flow time average calculation unit 24 reads the optical flow from the flow recording unit 23, calculates the absolute value of the optical flow for each of the x and y components (Sc3), and calculates the time average of the optical flow for a certain past time. (Sc4). Next, the blurring processing unit 25 blurs the pixel flow in the spatial direction (Sc5). The pixel flow here is the result of performing the time direction average and blurring processing (equivalent to the weighted average of neighboring pixels) on the absolute value of the calculated optical flow, and each pixel in the processing target area The pixel flow (vector) is calculated at (see the arrow in FIG. 8A). When this pixel flow is viewed as an image, as shown in FIG. 8B, the larger the absolute value, the higher the value (brighter).

Next, the flow statistic calculation unit 26 calculates the spatial inclination of the pixel flow in the processing target area subjected to the blurring process (Sc6). The spatial inclination is detected as a rapid change in the spatial direction of the size of the pixel flow in the processing target region by performing a differential filter (high-pass filter) process such as a Sobel filter. The output value of the Sobel filter is converted to an absolute value. Here, FIG. 8C shows the inclination in the x direction: dx, the inclination in the y direction: dy, and the absolute value √ (dx ² + dy ² ) of the pixel flow. In addition to the Sobel filter, there are various inclination detection filters, but the detection filter is not limited to what type.

  Next, the inclination of the pixel flow is obtained within the processing target area by the threshold comparison unit 28, and whether the inclination of the pixel flow is large according to the threshold specified by the threshold specification unit 27, that is, whether there is a value exceeding the threshold It is determined whether or not (Sc7), and the evaluation result is transmitted to the abnormality detection unit 29. Next, when the inclination of the pixel flow is large according to the evaluation result, the abnormality detection unit 29 detects that the pixel flow is abnormal (Sc8), determines whether the end condition is satisfied (Sc9), and ends the process. For example, the process returns to step Sc1, the target frame is advanced to the next frame, and the above-described processing is repeated.

  On the other hand, if the difference between the two is small, it is determined whether or not the end condition is satisfied without detecting an abnormality (Sc9). If not, the process returns to step Sc1 to advance the target frame to the next frame. The above process is repeated. Further, when the termination condition is satisfied, the process is terminated.

  In the above-described embodiment, the size of the filter for calculating the spatial gradient of the pixel flow needs to be matched with the size of the event to be detected. For example, if an abnormality occurs in a range of about 100 pixels, the filter needs to be able to detect its size. The simplest is to reduce the image size so that the sensitivity (size) of the filter and the spatial size of the event are roughly matched.

  In the above-described embodiment, the optical flow calculation unit 2 (21), the flow time average calculation unit 5 (24), the block average calculation unit 6, the average flow comparison unit 10, the abnormality detection unit 11 (29), the flow statistics The quantity calculation unit 26, the threshold comparison unit 28, and the like are executed in the computer system. The series of processes by the control unit 1, the image processing unit 5, and the calculation unit 6 are stored in a computer-readable recording medium in the form of a program, and the computer reads and executes the program. Thus, the above processing is performed. That is, optical flow calculation unit 2 (21), flow time average calculation unit 5 (24), block average calculation unit 6, average flow comparison unit 10, abnormality detection unit 11 (29), flow statistic calculation unit 26, threshold comparison The unit 28 is realized when a central processing unit such as a CPU reads the above program into a main storage device such as a ROM or RAM and executes information processing / arithmetic processing.

  Here, the computer-readable recording medium means a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD-ROM, a semiconductor memory, or the like. Alternatively, the computer program may be distributed to the computer via a communication line, and the computer that has received the distribution may execute the program.

It is a block diagram which shows the schematic structure of the abnormality detection apparatus of 1st Embodiment of this invention. It is a flowchart for demonstrating operation | movement of the abnormality detection apparatus by this 1st Embodiment. It is a conceptual diagram for demonstrating operation | movement of the abnormality detection apparatus by this 1st Embodiment. It is a block diagram which shows the structure of the abnormality detection apparatus by this 2nd Embodiment. It is a flowchart for demonstrating operation | movement of the abnormality detection apparatus by this 2nd Embodiment. It is a conceptual diagram for demonstrating operation | movement of the abnormality detection apparatus by this 2nd Embodiment. It is a flowchart for demonstrating operation | movement of the abnormality detection apparatus by this 3rd Embodiment. It is a conceptual diagram for demonstrating operation | movement of the abnormality detection apparatus by this 3rd Embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Block designation | designated part 2,21 ... Optical flow calculation part 3,22 ... Video data 4,23 ... Flow recording part 5,24 ... Flow time average calculation part 6 ... Block average calculation part 7,8 ... Average flow 9,27 ... threshold designation unit 10 ... average flow comparison unit 11, 29 ... abnormality detection unit (abnormality detection means)
20 ... Processing target area designating unit 25 ... Blur processing unit 26 ... Flow statistic calculation unit 28 ... Threshold comparison unit 100a ... Pixel flow calculation means 101a ... State index calculation means 100b ... Pixel flow calculation means 101b ... State index calculation means

Claims (8)

An input means for inputting a moving image to be monitored for each frame image;
A pixel flow calculation means for obtaining an optical flow in each image in a target region set in advance in the frame image, and calculating a pixel flow indicating the movement of each pixel from the optical flow;
State index calculation means for calculating a state index indicating the characteristics of the spatial distribution of the pixel flow in the frame image from the calculated pixel flow;
An abnormality detection device comprising: an abnormality detection unit that detects an abnormality of the monitoring target based on the state index. Provide at least two target areas in one frame image,
The state index calculating means includes
Based on each pixel flow in the target area, calculate the average flow of each target area,
The abnormality detection device according to claim 1, wherein the state index is calculated based on a difference in average flow of each target region. The state index calculating means includes
The abnormality detection device according to claim 2, wherein the state index is calculated based on a variance of each pixel flow in the target region. The state index calculating means includes
The abnormality detection device according to claim 2, wherein the state index is calculated based on a spatial change in each pixel flow in the target region. The pixel flow calculation means includes
5. The abnormality detection device according to claim 1, wherein an average of the optical flow of each pixel in a past predetermined time from the current frame image is calculated as each pixel flow. The pixel flow calculation means includes
6. The abnormality detection device according to claim 1, wherein each pixel flow is calculated based on an absolute value of an optical flow of each pixel. Input a moving image to be monitored for each frame image,
In a target region set in advance in the frame image, an optical flow in each image is obtained, and a pixel flow indicating movement of each pixel is calculated from the optical flow,
From the calculated pixel flow, a state index indicating the characteristics of the spatial distribution of the pixel flow in the frame image is calculated,
An abnormality detection method, comprising: detecting an abnormality of the monitoring target based on the state index. Inputting a moving image to be monitored for each frame image;
Obtaining an optical flow in each image in a target region set in advance in a frame image, and calculating a pixel flow indicating movement of each pixel from the optical flow;
Calculating a state index indicating the characteristics of the spatial distribution of the pixel flow in the frame image from the calculated pixel flow;
An abnormality detection program that causes a computer to execute the step of detecting an abnormality of the monitoring target based on the state index.

Images