JP4677561B2 - Water level detection system, water level detection method, and program - Google Patents

Description

  The present invention relates to a water level detection system, a water level detection method, and a program, and more particularly, to a water level detection technique for a river or the like based on a video signal obtained by photographing.

  The Ministry of Land, Infrastructure, Transport and Tourism has established a system that installs water level meters and monitoring cameras at various locations on the riverbed and collects the data obtained from them in one place in order to detect flood damage caused by river floods in advance and use it for flood control activities. Many telemeters have been implemented. In addition, a water level plate with an inclined pattern is installed in the river, and the water level of the river is detected from the river image without installing a water level meter by capturing the image including the level water plate and performing image processing. A method has been proposed (for example, see Patent Document 1).

Japanese Patent Laid-Open No. 9-161076

  However, the installation of objects that obstruct running water in rivers can cause floods and other disasters, so they are strictly controlled by the River Law, and there are only a limited number of places where water can be installed in rivers. . On the other hand, collecting water level data from as many locations as possible is indispensable for generating more accurate and reliable disaster prevention information.

  An object of the present invention is to make it possible to detect a water level of a river or the like from only a video signal having the river or the like as a subject without installing any object in the water.

The water level detection system of the present invention is a water level detection system that detects a water level based on an image captured using a fixed camera, and an image including a flowing region and a non-flowing region captured using a fixed camera. An integration means for integrating the supplied video in the time direction, a conversion means for performing wavelet transformation on the video obtained by the integration means, and a feature amount obtained by wavelet transformation in the conversion means , An estimation means for estimating for each pixel whether the pixel is the pixel of the flowing water region or the pixel of the non-flowing water region, and based on the estimation result of the estimating means, the in-video region is defined as the flowing water region and the non-flowing region. Area dividing means for detecting a water level by dividing into a flowing water area, and the area dividing means is estimated as a pixel estimated as the flowing water area or the non-flowing water area. Using the pixel as the target pixel, the number of the target pixel is integrated in the direction in which the boundary between the flowing water region and the non-flowing region extends, and the boundary pixel number is determined from the obtained distribution of the target pixel number. A boundary between the flowing water region and the non-flowing water region is detected as a water level by dividing the flowing water region and the non-flowing water region based on the obtained number of target pixels.
The water level detection method of the present invention is a water level detection method for detecting a water level based on an image captured using a fixed camera, and an image including a flowing water region and a non-flowing region captured using a fixed camera. An integration step for integrating in the time direction, a conversion step for performing wavelet transformation on the video image obtained in the integration step, and a pixel in the flowing water region based on the feature quantity obtained by the wavelet transformation or the non- An area for detecting a water level by dividing an in-video area into the running water area and the non-flowing water area based on an estimation process for estimating whether the pixel is a flowing water area pixel by pixel and an estimation result in the estimation process A division step, and in the region division step, the pixel estimated as the flowing region or the pixel estimated as the non-flowing region is set as a target pixel, and the number of target pixels is determined as the flowing water. Based on the obtained number of target pixels, with the boundary pixel number as a threshold value. A boundary between the flowing water region and the non-flowing water region is detected as a water level by dividing the flowing water region and the non-flowing water region.
The program of the present invention is a program for causing a computer to execute a water level detection process based on an image photographed using a fixed camera, and an image including a flowing water area and a non-flowing water area photographed using a fixed camera. Integration step in the time direction, a conversion step for performing wavelet transform on the video obtained in the integration step, and a pixel in the flowing water region based on the feature amount obtained by the wavelet transform or the above Based on the estimation step for estimating whether each pixel is a non-flowing region pixel and the estimation result in the estimation step, the water level is detected by dividing the in-video region into the flowing region and the non-flowing region. The area dividing step is executed by a computer, and the pixels estimated as the flowing water area or the top in the area dividing step are Using the pixel estimated as a non-flowing region as the target pixel, the number of target pixels is integrated in the direction in which the boundary between the flowing region and the non-flowing region extends, and the boundary pixel number is determined from the distribution of the obtained target pixels And detecting the boundary between the flowing water region and the non-flowing water region as a water level by dividing the flowing water region and the non-flowing water region based on the obtained target pixel number with the boundary pixel number as a threshold value. It is characterized by.

  According to the present invention, an image including a flowing region and a non-flowing region captured by a fixed camera with a river or the like as a subject is integrated in the time direction to suppress edge components in the flowing region, and wavelet transform is performed on the entire image. Based on the feature amount obtained by applying, it is estimated for each pixel whether it is a pixel in a flowing water region or a pixel in a non-flowing water region. And the boundary of a flowing water area | region and a non-flowing water area | region is detected as a water level by dividing | segmenting into a flowing water area | region and a non-flowing water area | region based on the estimation result. Thereby, the water level can be detected only from the video signal having the river or the like as the subject without installing any object in the water.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram illustrating a configuration example of a flowing water region detection system according to an embodiment of the present invention. The flowing water region detection system according to the present embodiment uses time-direction integration (for example, frame addition in the present embodiment), wavelet transform, and maximum likelihood estimation from a video signal having a subject such as a river obtained by photographing using a fixed camera CM. Based on the above, the flowing water area in the river or the like is detected.

  As shown in FIG. 1, the flowing water region detection system according to the present embodiment includes a region extraction unit 1, a frame addition unit 2, a wavelet transform unit 3, a maximum likelihood estimation unit 4, and a region division unit 5. Hereinafter, each function part of the flowing water area detection system shown in FIG. 1 will be described in detail. In the following, a case where the water level of a river is detected by detecting and discriminating a flowing water region and a non-flowing water region in an actual river image as shown in FIG.

(Region extraction unit 1)
The region extraction unit 1 extracts a region in which a land portion is captured in half and a flowing water portion in half, from a captured image captured using the fixed camera CM over several frames. Here, the flowing water part is an area where water such as a water surface in the captured image is reflected, and the land part is an area other than the flowing water part (for example, an area such as a background image).

  For example, the region extraction unit 1 extracts a region image in which the upper half is a land portion and the lower half is a flowing portion as shown in FIG. When extracting the area image, the land part may be below and the flowing part may be above, or the land part and flowing part may be shown on the left and right, or on the right and left, respectively, and the flowing area can be detected. . Further, it is also possible to perform shooting by adjusting the angle of the fixed camera CM so that half of the entire shot video is a land portion and the other half is a flowing water portion, and the obtained shot video itself may be used as a region image.

  Here, it is desirable that a stationary object with a pattern appears in the land portion in the region image, and that the water surface always moves in the flowing water portion. Further, it is desirable that the water surface is horizontal in the region image, and it is also effective to perform affine transformation such as rotating the photographed image (region image) 90 degrees or enlarging or reducing the image so as to be so. In addition, for a video taken at night, for example, as long as a certain pattern is reflected in the land as a result of infrared irradiation, nothing may be reflected in the flowing water, and the flowing water area can be detected.

  The region image is extracted after the user designates the extraction position at a certain point in time so that the water surface is always located near the center of the region image according to the water level obtained based on the detection result of the flowing water region. It may be performed automatically at time intervals.

(Frame adder 2)
The frame addition unit 2 synchronously adds a plurality of frames for the region image extracted by the region extraction unit 1. That is, the frame addition unit 2 integrates the region image extracted by the region extraction unit 1 in the time direction. As a result, an image in which the moving flowing water part is blurred or smooth is obtained. That is, the edge component (high frequency component) of the flowing water part is suppressed. On the other hand, the stationary land is not particularly changed. This can be confirmed, for example, by comparing FIG. 2 (a) and FIG. 2 (b). FIG. 2B is obtained by synchronously adding a plurality of frames by the frame adder 2.

  However, when adding region images of a plurality of frames, if the addition is simply performed, the pixel value simply increases and the required resources increase. For example, the pixel value is divided by the number of added frames. Therefore, it is desirable to prevent the range of pixel values from expanding. Further, for example, when adding images, the pixel value may be multiplied by a weighting factor, or a bit shift may be performed on the pixel value.

  For example, when adding four frames, the frame adding unit 2 multiplies the pixel value in each frame by 0.25 or shifts the pixel value to the right by 2 bits so as to be equivalent to it. The pixel value already stored in the frame memory is added, and the addition result is newly stored in the frame memory. Thus, frame addition can be realized with a small number of frame memories without preparing as many frame memories as the number of sheets to be added.

  Further, the pixel value stored in the frame memory and the pixel value of the region image extracted by the region extraction unit 1 are added and divided by 2, and then the addition result is newly stored in the frame memory. By repeating this process, an image is generated by weight-adding region images of a plurality of frames. Alternatively, when taking an image with the fixed camera CM, an effect equivalent to that obtained by adding a plurality of frames as described above can be obtained by opening the shutter for a relatively long time and making the exposure time relatively long. be able to.

  Here, although details will be described later, in the present embodiment, the land portion and the flowing water portion are determined by setting a region including many edge pixels in the video as a land portion and a region including few edge pixels as a flowing portion. Then, by recognizing the position of the boundary between the land portion and the flowing water portion as the water surface, it becomes possible to calculate the water level as shown in FIG.

(Wavelet transform unit 3)
The wavelet transform unit 3 performs wavelet transform on the image generated by the frame addition unit 2 and emphasizes the edge component that is small in the flowing part but large in the land part, that is, the difference between the flowing part and the land part. Such N-dimensional (N is an integer) feature vector (feature amount) is calculated. For example, by performing wavelet transform by the wavelet transform unit 3 on the image shown in FIG. 2B, four band images such as LL, LH, HL, and HH are generated as shown in FIG. 2C. . However, the number of pixels in the band image is halved both vertically and horizontally before the wavelet transform.

  These four band images (band signals) may be the four components of the feature vector for each pixel of the image generated by the frame addition unit 2. That is, the feature vector shown in FIG. 2C is a four-dimensional vector. However, since the number of pixels is halved both vertically and horizontally as described above, a group of four adjacent pixels in the image generated by the frame adder 2 corresponds to one pixel in the band image. Therefore, a feature vector is generated with four pixels as one group.

  Further, the wavelet transform may be applied once more to the LL band image (low frequency image) shown in FIG. 3C, and in this case, seven band images are generated in total. The feature vector is a 7-dimensional vector. Note that it is not always necessary to use all of the band images obtained by the wavelet transform, and band images in which the difference between the land part and the flowing water part is not seen may not be included in the feature vector.

  There are many types of wavelet transform bases, and any of them may be used. In particular, by using wavelet transform in JPEG2000, which is an international standard for image compression (see, for example, “JPEG 2000-Image compression fundamentals, standards and practice”, co-authored by DSTaubman and MWMarcellin, Kluwer Academic Publishers, 2002) A general-purpose IP core can be used to shorten the system development period and reduce the circuit size.

(Maximum likelihood estimator 4)
The maximum likelihood estimation unit 4 estimates whether each pixel in the image generated by the wavelet transform unit 3 is a land region pixel or a flowing water region pixel. Further, for example, the maximum likelihood estimation unit 4 changes the pixel value to either white (a pixel estimated as a land area) or black (a pixel estimated as a flowing water area) according to the estimation result, and FIG. An image as shown in FIG.

When estimating whether the pixel is a land region pixel or a flowing water region pixel, for example, a maximum likelihood estimation method can be used. When the maximum likelihood estimation method is used, first, image areas that are already known to be a running water area and a land area are designated as teacher images. Here, these are called class ω ₁ and class ω ₂ , respectively. Next, with respect to an image to be discriminated, it is discriminated which class belongs to each pixel.

As a result of the maximum likelihood estimation, a class ω _i having a maximum prior probability P (G (m, n) | ω _i ), iε {1,2} is determined for each pixel. If the probability density distribution of the feature vector values can be assumed to be a normal distribution, the class that minimizes the log-likelihood function of Determined.

Here, μ _i and C _i are the mean value vector and covariance matrix of feature vectors obtained from the class ω _i teacher image, and d _i ² is the Mahalanobis distance.

  In the above description, one class is assigned to each of the flowing water area and the land area, but a plurality of classes can be assigned. In addition, the position of the teacher image is automatically set on the basis of the water level that is updated from time to time, such as the flowing water portion below the already calculated water level and the land portion above the water level, Stable running water area detection is possible against changes in sunshine and water level.

(Region division unit 5)
The region dividing unit 5 discriminates, from the image generated by the maximum likelihood estimating unit 4, a region including many pixels estimated to be a land region as a land portion and the other as a flowing portion, and a non-flowing region. And divided into two areas. This process is performed based on the pixel distribution estimated by the maximum likelihood estimation unit 4 as a land area or the pixel distribution estimated as a flowing water area.

  For example, by outputting the boundary between the two regions thus obtained, the water level is calculated as the boundary between the land portion and the flowing water portion as shown in FIG.

  Various forms of processing in the area dividing unit 5 are conceivable. Hereinafter, an example of processing in the region dividing unit 5 will be described with reference to FIGS.

  FIG. 3 is a block diagram showing a configuration example of the area dividing unit 5, and FIG. 4 is a diagram for explaining the principle of area division by the area dividing unit 5 shown in FIG.

(Horizontal histogram calculation unit 11)
The horizontal histogram calculation unit 11 determines how many pixels (hereinafter also referred to as “estimated land portion pixels”) that are estimated to be land regions by the maximum likelihood estimation unit 4 in the target region image. Count for each line. That is, the horizontal direction histogram calculation unit 11 integrates the estimated number of land pixels for each row and generates a horizontal histogram of the estimated land pixels.

  For example, in the binary image schematically shown in FIG. 4A, there are six estimated land pixels in the first row from the top, and seven estimated land pixels in the second row. Accordingly, in the horizontal histogram (FIG. 4B) generated by the horizontal histogram calculation unit 11, the estimated land pixel numbers are plotted as 6 and 7 for each row, that is, the vertical position.

(Vertical Histogram Calculation Unit 12)
The vertical histogram calculation unit 12 refers to the horizontal histogram generated by the horizontal histogram calculation unit 11 and counts how many corresponding vertical positions exist for each estimated land pixel number value. That is, the vertical direction histogram calculation unit 12 generates a vertical direction histogram regarding each estimated land portion pixel number in the horizontal direction histogram.

  For example, in the horizontal histogram shown in FIG. 4B, there are two estimated land pixel numbers in the first column from the left, and one estimated land pixel number in the second column. In the horizontal histogram (FIG. 4C) generated by the vertical histogram calculation unit 12, the vertical integrated values of the estimated land pixel numbers are 2 and 1 for each column, that is, the estimated land pixel number. Is plotted as

(Boundary value determination unit 13)
The vertical histogram generated by the vertical histogram calculation unit 12 includes two clusters (distributions) of a histogram portion related to the land portion and a histogram portion related to the flowing water portion. Therefore, it is possible to determine the boundary portion that separates these two clusters. The boundary value determination unit 13 determines the boundary portion (number of boundary pixels). Various methods can be used as a method for determining the boundary portion.

  For example, the boundary value determination unit 13 may determine the boundary so that the variance within the class is small and the distance between the classes is large, or the boundary so that the intra-class variance / inter-class variance ratio is maximized. May be determined. Alternatively, the boundary value determination unit 13 may determine the boundary between the two classes by reducing the vertical histogram to a bimodal Gaussian distribution.

(Water level calculation unit 14)
Based on the two class boundaries in the vertical histogram determined by the boundary value determination unit 13, the water level calculation unit 14 determines the boundary in the corresponding horizontal histogram, and further determines the boundary of the corresponding binary image. . As a result, the image is divided into two regions, a land portion and a flowing water portion, and the boundary is output as a water level.

  Specifically, with the boundary determined by the boundary value determination unit 13 as a threshold value, the corresponding boundary P1 in the vertical histogram shown in FIG. 4C is the horizontal direction shown in FIG. 4B. Through the corresponding point P2 in the histogram, a water level P3 that is a boundary between the land portion and the flowing water portion in the binary image shown in FIG. The case where it applies to an actual river image is shown in FIG.

  In the above description, the case where the water surface is horizontal in the image (the boundary between the land portion and the flowing water portion extends in the lateral direction) is shown as an example, but the water surface is vertical in the image (the land portion and the flowing water). When the boundary of the portion extends in the vertical direction), the edge pixel number is added up for each column to generate a vertical histogram of the edge pixel, and the horizontal direction of each edge pixel number is referenced with reference to that. A histogram may be generated.

  As described above, according to the present embodiment, from the captured image captured using the fixed camera CM with a river or the like as a subject, the region extraction unit 1 uses the boundary between the land part and the flowing part, that is, the land part and the flowing part. Is extracted, and the frame adding unit 2 adds several frames for each region to generate an image in which the running water portion is smooth. Further, the wavelet transform is performed on the entire image by the wavelet transform unit 3, and the maximum likelihood estimation unit 4 estimates whether the pixel is a land pixel or a flowing water pixel based on the feature vector obtained by the wavelet transform. Based on the estimation result, the region dividing unit 5 performs region division on the land portion and the flowing water portion. As a result, the land part and the flowing water part can be detected, respectively, and the boundary between the two areas is determined as the water level, so that the flowing water can be detected only from the video signal with the river or the like as the subject without installing any object in the water. A region can be detected.

  In the above description, the area extraction unit 1 extracts one area image from the captured video imaged using the fixed camera CM. However, as shown in FIG. On the other hand, a plurality of region images (region 1, region 2, and region 3 in the example shown in FIG. 6) may be extracted. In that case, the partial areas may overlap in the plurality of area images. For each of these area images, the land part and the flowing water part are discriminated by the above-described processing, and the final division of the land part and the flowing water part is performed by comprehensively judging from the obtained results. May be. For example, rainwater adheres to the lens of the fixed camera CM by using the average value of multiple results as the boundary between the final land part and the flowing water part, or by determining the final land part and flowing water area by the majority method. Even when an image failure due to rainfall, snowfall or fog occurs and it is difficult to detect the flowing water portion, the flowing water portion can be detected with high accuracy and stability.

(Other embodiments of the present invention)
The above-described embodiment can be configured by a CPU or MPU of a computer, a RAM, a ROM, and the like, and can be realized by operating a program stored in the RAM or the ROM, and the program is an embodiment of the present invention. include. In addition, a program that causes a computer to operate so as to perform the functions of the above-described embodiments can be realized by recording the program on a recording medium such as a CD-ROM and causing the computer to read the program. A medium is included in an embodiment of the present invention. As a recording medium for recording the program, a flexible disk, a hard disk, a magnetic tape, a magneto-optical disk, a nonvolatile memory card, and the like can be used in addition to the CD-ROM.
In addition, a program product in which the functions of the above-described embodiments are realized by a computer executing a program and performing processing is included in the embodiments of the present invention. The program product includes the program itself that implements the functions of the above-described embodiments, a computer loaded with the program, a transmission device that can provide the program to a computer that is communicably connected via a network, and the transmission There are network systems equipped with devices.
Moreover, not only the functions of the above-described embodiments are realized by executing a program supplied by a computer, but the program is also shared with an OS (operating system) or other application software running on the computer. When the functions of the above-described embodiment are realized, or when all or part of the processing of the supplied program is performed by a function expansion board or a function expansion unit of the computer, the functions of the above-described embodiment are realized. Such a program is included in the embodiment of the present invention.

For example, a computer function 700 as shown in FIG. 7 is provided, and the CPU 701 performs the operation in the above-described embodiment.
As shown in FIG. 7, the computer function 700 includes a CPU 701, a ROM 702, a RAM 703, a keyboard controller (KBC) 705 of a keyboard (KB) 709, and a CRT controller (CRTC) of a CRT display (CRT) 710 as a display unit. ) 706, a disk controller (DKC) 707 of a hard disk (HD) 711 and a flexible disk (FD) 712, and a network interface card (NIC) 708 are communicably connected to each other via a system bus 704. Yes.
The CPU 701 comprehensively controls each component connected to the system bus 704 by executing software stored in the ROM 702 or the HD 711 or software supplied from the FD 712.
That is, the CPU 701 reads out a processing program for performing the above-described operation from the ROM 702, the HD 711, or the FD 712 and executes it, thereby performing control for realizing the operation in the above-described embodiment.
The RAM 703 functions as a main memory or work area for the CPU 701.
The KBC 705 controls an instruction input from the KB 709 or a pointing device (not shown).
The CRTC 706 controls display on the CRT 710.
The DKC 707 controls access to the HD 711 and the FD 712 that store a boot program, various applications, user files, a network management program, the processing program, and the like.
The NIC 708 exchanges data with other devices on the network 713 in both directions.

  The above-described embodiments are merely examples of implementation in carrying out the present invention, and the technical scope of the present invention should not be construed as being limited thereto. That is, the present invention can be implemented in various forms without departing from the technical idea or the main features thereof.

It is a figure which shows the structural example of the flowing water area | region detection system by one Embodiment of this invention. It is a figure which shows an example of the image processed with the flowing water area | region detection system shown in FIG. It is a figure which shows the structural example of an area division part. It is a figure for demonstrating the principle of the area division by the area division part shown in FIG. It is a figure which shows the case where the area division by the area division part shown in FIG. 3 is applied to an actual river image. It is a figure which shows the other example of the area image extraction by an area extraction part. It is a block diagram which shows the computer function which can implement | achieve the flowing water area | region detection system in this embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Area extraction part 2 Frame addition part 3 Wavelet transformation part 4 Maximum likelihood estimation part 5 Area division part 11 Horizontal direction histogram calculation part 12 Vertical direction histogram calculation part 13 Boundary value determination part 14 Water level calculation part

Claims (7)

A water level detection system that detects a water level based on an image captured using a fixed camera,
An integration unit that integrates the supplied image in the time direction, provided with an image including a flowing region and a non-flowing region captured using a fixed camera;
Conversion means for performing wavelet transform on the video obtained by the integration means;
An estimation unit that estimates, for each pixel, whether the pixel is in the flowing region or the non-flowing region based on the feature amount obtained by the wavelet transform in the conversion unit;
An area dividing means for detecting a water level by dividing an in-video area into the flowing water area and the non-flowing water area based on an estimation result in the estimating means;
The region dividing means integrates the number of target pixels with respect to a direction in which a boundary between the flowing water region and the non-flowing region extends, with the pixel estimated as the flowing water region or the pixel estimated as the non-flowing water region as a target pixel. Further, the boundary pixel number is determined from the distribution of the obtained target pixel number, and the boundary pixel number is used as a threshold value to divide the flowing water region and the non-flowing region based on the obtained target pixel number. A water level detection system for detecting a boundary between a flowing water region and the non-flowing water region as a water level.   2. The water level detection system according to claim 1, wherein the integration means adds the supplied video over a plurality of frames.   The said estimation means estimates whether it is a pixel of the said water flow area | region or the pixel of the said non-flow water area | region, using the several band signal divided | segmented by the said wavelet transformation as the said feature-value. The water level detection system according to 1 or 2.   The apparatus according to claim 1, further comprising a region extracting unit that extracts a region image including the flowing region and the non-flowing region from an image captured using the fixed camera and supplies the region image to the integrating unit. The water level detection system according to any one of claims.   Extracting the plurality of region images having different regions from the image captured using the fixed camera, and processing each of the region images obtained by processing each extracted region image 5. The water level detection system according to claim 4, wherein a water level in the photographed video is detected based on the division result. A water level detection method for detecting a water level based on an image taken with a fixed camera,
An integration step of integrating in a time direction an image including a flowing water region and a non-flowing region photographed using a fixed camera;
A conversion step of performing wavelet conversion on the video obtained in the integration step;
An estimation step for estimating, for each pixel, whether the pixel is in the flowing region or the non-flowing region based on the feature amount obtained by the wavelet transform;
A region dividing step of detecting a water level by dividing a region in the image into the flowing water region and the non-flowing water region based on the estimation result in the estimating step;
In the region dividing step, the pixel estimated as the flowing region or the pixel estimated as the non-flowing region is set as the target pixel, and the number of target pixels is integrated in the direction in which the boundary between the flowing region and the non-flowing region extends. Further, by determining the number of boundary pixels from the obtained distribution of the number of target pixels, and dividing the divided into the flowing region and the non-flowing region based on the obtained number of target pixels using the boundary pixel number as a threshold value. A water level detection method, wherein a boundary between the flowing water region and the non-flowing water region is detected as a water level. A program for causing a computer to execute a water level detection process based on an image captured using a fixed camera,
An integration step of integrating in a time direction an image including a flowing water region and a non-flowing region photographed using a fixed camera;
A conversion step of performing wavelet transform on the video obtained in the integration step;
An estimation step for estimating, for each pixel, whether the pixel is the pixel of the flowing water region or the pixel of the non-flowing region based on the feature amount obtained by the wavelet transform;
Based on the estimation result in the estimation step, the computer executes an area division step of detecting a water level by dividing an in-video area into the flowing water area and the non-flowing area,
In the region dividing step, the pixel estimated as the flowing region or the pixel estimated as the non-flowing region is set as the target pixel, and the number of target pixels is integrated in the direction in which the boundary between the flowing region and the non-flowing region extends. Further, by determining the number of boundary pixels from the obtained distribution of the number of target pixels, and dividing the divided into the flowing region and the non-flowing region based on the obtained number of target pixels using the boundary pixel number as a threshold value. A program for detecting a boundary between the flowing water region and the non-flowing water region as a water level.