JP2014164525A - Method, device and program for estimating number of object - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method, device, and program for estimating the number of objects capable of accurately estimating the number of objects which exist in a specific region on the basis of a photographic image photographed at a plurality of photographic positions.SOLUTION: A foreground image extraction part 30 extracts an image from a photographic image and a background image. An object existable region calculation part 32 calculates a sum x of volumes in an object existable region by a visual volume intersection method on the basis of the position and angle parameter of each camera and each foreground image. An object number estimation value calculation part 34 calculates and outputs the estimation value of the number of objects by using the parameter of regression analysis with the sum x of the volumes in the object existable region as explanatory variables and the number of objects y as objective variables. The parameter of the regression analysis to be used by the object number estimation value calculation part 34 is determined in advance by an object number estimation parameter determination part 44 of a parameter learning part 14 on the basis of learning data.

Description

  The present invention relates to an object number estimation method, an object number estimation device, and an object number estimation program, and in particular, the number of objects for estimating the number of objects in a specific region based on captured images captured at a plurality of imaging positions. The present invention relates to an estimation method, an object number estimation device, and an object number estimation program.

  In recent years, when measuring the effects of digital signage and advertising, ascertaining the number of visitors at exhibitions, analyzing surveillance camera images, etc., it is important to use techniques to calculate the number of people in a specific area from images captured by cameras. It has become a thing.

  For example, a technique described in Non-Patent Document 1 is known as a method of estimating the number of persons existing in a specific area such as a room using an image captured by a camera. In this Non-Patent Document 1, a technique for calculating the number of persons in a specific region based on the area occupied by each pixel on the surface of the person in real space using an image taken by one camera. Is described.

Hiroyuki Arai, Isao Miyagawa, Hideki Koike, Miki Haseyama, “Estimating the number of people from images based on geometric models”, Technical Report of the Institute of Image Information and Television Engineers, Vol. 32, No. 26, pp. 33-36.

  However, in the technique described in Non-Patent Document 1, when a plurality of persons are arranged on a straight line extending from the camera, it is difficult to count the persons hidden behind the frontmost person. (See FIG. 3).

  The present invention has been made in view of the above circumstances, and can estimate the number of objects that can accurately estimate the number of objects existing in a specific area based on captured images taken at a plurality of shooting positions. It is an object to provide a method, an object number estimation device, and an object number estimation program.

  In order to achieve the above object, the object number estimation method according to the present invention includes a plurality of background images obtained by photographing a specific region from a plurality of photographing positions in a state where no object exists in advance by a photographing foreground image extracting unit, A step of extracting a foreground image, which is an image representing an object, from each of the plurality of photographed images based on a difference between the plurality of photographed images obtained by photographing the specific region from the photographing position; A region where an object can exist in the specific region obtained by using a visual volume intersection method based on a plurality of photographing conditions including the plurality of photographing positions and a plurality of the foreground images extracted by the photographing foreground image extracting unit. And calculating the volume of each of the plurality of the background images and the specific region where the objects having a known number of objects exist by the object number estimating means. Parameters of regression analysis, which are obtained in advance based on a plurality of learning images photographed at a position, with the sum of the volumes of the regions where the objects can exist as explanatory variables and the number of known objects as an objective variable And calculating and outputting an estimated value of the number of objects in the specific area based on the sum of the volumes of the possible areas of the objects calculated by the object possible area calculation means. .

  In the object number estimation method of the present invention, the learning foreground image extracting means uses a plurality of learning images in which the plurality of background images and the specific area where objects with a known number of objects exist are photographed at the plurality of photographing positions. Extracting a learning foreground image, which is an image representing an object, for each of the plurality of learning images based on the difference between the image and the image for learning, and the learning object existence possible area calculation means, Based on the learning foreground image, the step of calculating each volume of the object possible area in the specific area obtained by using the visual volume intersection method, and the learning object existence possible area calculation by the parameter determining means And determining a regression analysis parameter using the sum of the volume of the object possible area calculated by the means as an explanatory variable and the number of known objects as an objective variable. Wherein the object number estimating means uses the parameter parameter determining means has determined, it is preferable to perform the regression analysis calculates and outputs an estimate of the number of objects in the specific region.

  Further, the object number estimation method of the present invention includes a step of calculating a predetermined image feature for the learning image and the learning foreground image by a learning image feature calculation unit, and an image feature calculation unit for calculating the captured image and the Calculating a predetermined image feature for a foreground image, and the parameter determining means includes a sum of the volumes of the object existable areas calculated by the learning object existable area calculating means and the Estimating the number of objects by determining parameters for regression analysis using the predetermined image features calculated for the learning image and the learning foreground image by the learning image feature calculation means as explanatory variables and the number of known objects as an objective variable And means for calculating a sum of volumes of the object existable areas calculated by the object existable area calculating means and the image feature meter. Based on a predetermined image features calculated for the captured image and the foreground image by means may output to calculate an estimate of the number of objects.

  Further, in the estimation method of the present invention, the predetermined image feature calculated for the learning image and the learning foreground image includes an area of each of the plurality of learning foreground images and a predetermined value of each of the plurality of learning images. The predetermined image feature calculated for the captured image and the foreground image includes an area of each of the plurality of foreground images and the predetermined image feature amount of each of the plurality of captured images. It is preferable to include.

  The object number estimation device of the present invention is a plurality of background images obtained by photographing a specific area from a plurality of photographing positions in the absence of an object in advance, and a plurality of photographed images obtained by photographing the specific area from the plurality of photographing positions; A foreground image extracting means for extracting a foreground image, which is an image representing an object, from each of the plurality of photographed images, the photographing conditions including the plurality of photographing positions, and the photographing foreground image extracting means. An object existence possible area calculation means for calculating the volume of each of the object existence possible areas in the specific area obtained by using the visual volume intersection method based on the plurality of foreground images extracted by Presence of the object can be obtained, which is obtained in advance based on the background image and a plurality of learning images obtained by photographing the specific region in which a known number of objects are present at the plurality of photographing positions. The volume of each of the object existable areas calculated by the object existable area calculating means using the regression analysis parameters with the sum of the volume of each area as the explanatory variable and the known number of objects as the objective variable Object number estimating means for calculating and outputting an estimated value of the number of objects in the specific region based on the sum of

  The object number estimation program of the present invention is for causing a computer to execute each step of the object estimation method of the present invention.

  According to the object number estimation method, the object number estimation device, and the object number estimation program of the present invention, it is possible to accurately estimate the number of objects existing in a specific region based on captured images captured at a plurality of capturing positions. The effect of being able to be obtained.

It is a block diagram which shows the outline of an example of the object number estimation apparatus of 1st Embodiment. It is explanatory drawing which shows an example of arrangement | positioning of the camera in 1st Embodiment, and the object (person) which should be counted. FIG. 3 is a schematic diagram in a case where a single camera (only camera 1) is used to image the specific area shown in FIG. (A) is the figure which looked at the specific area from right above. (B) is a photographed image photographed by the camera 1. It is a schematic diagram in case the inside of the specific area shown in FIG. 2 is imaged with two cameras (camera 1 and camera 2). (A) is the figure which looked at the specific area from right above. (B) is a photographed image photographed by the camera 1. (C) is a photographed image photographed by the camera 2. It is a flowchart of an example of the object number estimation process performed in the object number estimation part of the object number estimation apparatus of 1st Embodiment. It is explanatory drawing showing the area | region where the object reflected on the picked-up image can exist when the picked-up image of the camera 1 and the picked-up image of the camera 2 are given. It is explanatory drawing for demonstrating the calculation of the sum of the volume of the object possible area | region using a visual volume intersection method in the general case where a several object exists. It is a flowchart of an example of the parameter learning process performed in the parameter learning part of the object number estimation apparatus of 1st Embodiment. It is a block diagram which shows the outline of an example of the object number estimation apparatus of 2nd Embodiment. It is a flowchart of an example of the object number estimation process performed in the object number estimation part of the object number estimation apparatus of 2nd Embodiment. It is a flowchart of an example of the parameter learning process performed in the parameter learning part of the object number estimation apparatus of 2nd Embodiment.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Note that this embodiment does not limit the present invention.
(First embodiment)
In the present embodiment, as a specific example, the number of objects to be estimated (number of objects) in a specific region using captured images captured by two cameras (so-called fixed point observation cameras) whose capturing positions are fixed. ) Will be described. In the present embodiment, “object” refers to both living things and non-living objects including a person, and a case where the object to be estimated is a person will be described as an example.

  First, a schematic configuration of the object number estimation apparatus 10 of the present embodiment will be described. The object number estimation apparatus 10 according to the present embodiment uses a computer including a CPU (Central Processing Unit), a RAM (Random Access Memory), and a ROM (Read Only Memory). When the CPU executes the object number estimation processing program stored in the ROM, the object number estimation processing and parameter learning processing, which will be described in detail later, are executed.

  FIG. 1 is a block diagram illustrating an example of a functional configuration of the object number estimation apparatus 10 according to the present embodiment. As shown in FIG. 1, the object number estimation device 10 of the present embodiment includes an object number estimation unit 12, a parameter learning unit 14, a background image storage unit 20, a camera position / angle parameter storage unit 22, and an object number estimation parameter. A storage unit 24 is provided.

  In the background image storage unit 20, background images obtained by photographing a specific area in a state where no object exists by each of the two cameras (camera 1 and camera 2) are stored in advance. In the present embodiment, the “state where no object exists” means a state where there is no object for which the number of objects is to be estimated, and is provided in a specific area where the number of objects is not to be estimated. It does not prevent the existence of furniture that exists.

  The camera position / angle parameter storage unit 22 stores in advance the position (shooting position) and angle parameters of the camera 1 and the camera 2 (more specifically, the lens of each camera). The camera position / angle parameter accumulating unit 22 only needs to accumulate data necessary for calculating an object existence possible area (details will be described later) by the view volume intersection method. And is not limited to the camera position / angle parameter.

  The object number estimation parameter accumulation unit 24 accumulates the object number estimation parameter determined by the object number estimation parameter determination unit 44 of the parameter learning unit 14.

  The object number estimation unit 12 includes a foreground image extraction unit 30, an object existence possible region calculation unit 32, and an object number estimation value calculation unit 34.

  The foreground image extraction unit 30 has a function of extracting a foreground image of a photographed image photographed by the camera 1 from a photographed image photographed by the camera 1 and a background image of the camera 1 accumulated in the background image accumulation unit 20. have. Further, the foreground image extraction unit 30 extracts a foreground image of the photographed image photographed by the camera 2 from the photographed image photographed by the camera 2 and the background image of the camera 2 accumulated in the background image accumulation unit 20. It has a function to do. In the present embodiment, the “foreground image” refers to an object image representing an object in a captured image. When a plurality of object images are included in the captured image, the foreground image includes all object images.

  The object existence possible area calculation unit 32 uses the view volume intersection method from the foreground image of the camera 1 and the foreground image of the camera 2 extracted by the foreground image extraction unit 30, and uses each of the object existence possibility areas in the specific area. It has a function of calculating the sum x of volumes.

  The object number estimated value calculation unit 34 is based on the sum x of the volume of the object existable region calculated by the object existable region calculation unit 32 and the regression analysis parameters stored in the object number estimation parameter storage unit 24. Thus, it has a function of calculating and outputting an estimated value of the number of objects according to the regression equation.

  The parameter learning unit 14 includes a learning foreground extraction unit 40, a learning object existence possible region calculation unit 42, and an object number estimation parameter determination unit 44. The parameter learning unit 14 sets parameters for estimating the number of objects using the learning image captured by the camera 1 and the learning image captured by the camera 2 in a state where an object having a known number of objects exists in the specific region. It has a function to learn (determine).

  The learning foreground extraction unit 40 extracts the foreground image of the learning image captured by the camera 1 from the learning image captured by the camera 1 and the background image of the camera 1 stored in the background image storage unit 20. It has a function to do. The learning foreground extraction unit 40 also uses the learning image captured by the camera 2 and the background image of the camera 2 stored in the background image storage unit 20 to foreground images of the learning image captured by the camera 2. Has the function of extracting.

  The learning object existence possible area calculation part 42 uses the view volume intersection method from the foreground image of the camera 1 and the foreground image of the camera 2 extracted by the learning foreground extraction part 40, and the volume of the object existence possibility area in the specific area. The function of calculating the sum x of

  The object number estimation parameter determination unit 44 uses the sum x of the volume of the object existence possible area calculated by the learning object existence possible area calculation unit 42 and the volume of each object existence possible area using the known number of objects. The function of determining the parameters of regression analysis with the sum x of x as the explanatory variable and the number of objects as the objective variable.

  Next, the object number estimation process executed by the object number estimation apparatus 10 of the present embodiment will be described. In the object number estimation process according to the present embodiment, as described above, the number of objects in the specific area is estimated based on the captured images captured by the two cameras. That is, the number of objects in the specific area is estimated based on captured images captured at different imaging positions.

  FIG. 2 shows an example of the arrangement of cameras and an object (person) to be counted in the present embodiment. In the example shown in FIG. 2, a person is present in a partitioned space such as a room, and the positions, viewing angles, and optical axis directions of the cameras 1 and 2 are fixed.

  Here, in the example shown in FIG. 2, consider a case where there is one camera and one shooting position. FIG. 3 is a schematic diagram in a case where the camera captures the inside of the specific area shown in FIG. 2 with one camera (only camera 1). FIG. 3A is a view of the specific region as viewed from directly above. FIG. 3B shows a captured image captured by the camera 1. The projection image on the projection plane in FIG. 3A corresponds to the captured image shown in FIG.

  As shown in FIG. 3 (A), actually, as shown in FIG. 3 (B), although the four objects 1 to 4 exist in the specific region, the captured image is displayed. There are only two object images. Since the objects 1 to 3 are arranged in a straight line extending from the camera 1, the objects 2 and 3 are hidden behind the object 1. Therefore, although there are three objects, only the image of the object 1 appears in the captured image. In this case, the foreground image includes only one object image.

  On the other hand, as in the present embodiment, FIG. 4 shows a schematic diagram in the case where two cameras (camera 1 and camera 2) are used to photograph the specific area shown in FIG. FIG. 4A is a view of the specific area as viewed from directly above. FIG. 4B shows a captured image captured by the camera 1. FIG. 4C shows a photographed image photographed by the camera 2. The projection image on the projection plane of the camera 1 in FIG. 4A corresponds to the captured image shown in FIG. In addition, the projection image on the projection plane of the camera 2 in FIG. 4A corresponds to the captured image shown in FIG.

  As shown in FIG. 4C, the captured image of the camera 2 shows four object images, and all object images of the objects 1 to 4 are shown. As described above, by using the photographed images taken from different photographing positions, the number of objects existing in the specific region is more accurately reflected in the photographed image. In general, since the object is not stationary and the position is indefinite, it cannot be determined which captured image of the camera 1 or the camera 2 is accurate. In addition, the captured images of both the camera 1 and the camera 2 may be inaccurate. However, by complementing the information from the camera 1 and the camera 2, the number of objects can be estimated more accurately.

  FIG. 5 shows a flowchart of an example of the object number estimation process executed by the object number estimation apparatus 10 of the present embodiment.

  The object number estimation process is executed, for example, when captured images are input from the camera 1 and the camera 2 to the object number estimation apparatus 10. Further, for example, the object number estimation apparatus 10 may acquire a captured image from the camera 1 and the camera 2 every predetermined time, and execute the object number estimation process. The execution timing is not particularly limited, and may be executed as desired by the user. In any case, when the CPU of the object number estimation apparatus 10 executes the object number estimation program stored in the ROM, the object number estimation process shown in FIG. 5 and the parameter learning process shown in FIG. 8 are executed. Is done.

  First, in step S100, the foreground image extraction unit 30 acquires a captured image input from the camera 1.

  In the next step S102, the foreground image extraction unit 30 acquires a background image corresponding to the camera 1 from the background image storage unit 20. As described above, in the foreground image extraction unit 30, the background image captured by the camera 1 in a state where no object exists in the specific area is stored in association with the camera 1 in advance.

  In the next step S104, the foreground image corresponding to the object image is extracted based on the photographed image taken by the camera 1 and the background image acquired from the foreground image extraction unit 30.

  In the present embodiment, the foreground image is extracted using the difference between the captured image and the background image. In the present embodiment, as an extraction method, an object image area in the captured image is identified by detecting pixels with a large difference between the RGB value of the captured image and the RGB value of the background image, and the identified area is identified as the foreground image. As extracted.

  Note that the value used for the difference is not particularly limited, and for example, a general numerical value expressing the color space can be used in addition to the RGB value. As a difference calculation method, for example, a commonly used distance measure such as a norm of a vector difference when an RGB value is regarded as a vector can be used.

  In order to improve the accuracy of foreground image extraction, for example, literature (Shigeki Nagaya, Takatake Miyatake, Takehiro Fujita, Hiroyuki Ueda, Keiichi Ito, “Moving Object Detection by Time Correlation Background Judgment”, electronic information You may use the existing method described in IEICE Transactions Journal, J79-D-II (4), pp.568-576.

  Thus, the extraction method is not particularly limited, and an existing method may be used.

  In the next step S106, it is determined whether foreground images have been extracted from the captured images of all cameras. If not extracted, for example, in the present embodiment, if only the foreground image of the camera 1 is extracted, the process returns to step S100 to repeat this process and extract the foreground image from the captured image of the camera 2. On the other hand, if foreground images are extracted from the images captured by all cameras, the process proceeds to step S108. Instead of repeating the process of extracting the foreground for each camera in this way, the process of each step may be performed simultaneously on the captured images of all the cameras.

  In step S <b> 108, the object existence possible area calculation unit 32 acquires the positions (photographing positions) and angle parameters of the camera 1 and the camera 2 from the camera position / angle parameter storage unit 22. As described above, the camera position / angle parameter storage unit 22 stores the positions (shooting positions) and angle parameters of the cameras 1 and 2 in association with each other in advance.

  In the next step S110, the object existence possible area calculation unit 32 performs real space based on the foreground image of the camera 1 extracted by the foreground image extraction unit 30, the foreground image of the camera 2, and the camera position / angle parameters of each camera. Each of the object existable regions x is calculated using the visual volume intersection method, and the sum x of the volumes of the object existable regions is calculated.

  Here, the visual volume intersection method will be described with reference to FIGS. 6 and 7. FIG. 6 is an explanatory diagram showing an area where the object represented by the foreground image in the photographed image can exist when the photographed image of the camera 1 and the photographed image of the camera 2 are given.

  As shown in FIG. 6, when an object is photographed, the object image 1 of the object is reflected on the projection surface of the camera 1. Further, the object image 2 of the object appears on the projection surface of the camera 2. The projection surface of the camera 1 corresponds to the captured image of the camera 1, and the projection surface of the camera 2 corresponds to the captured image of the camera 2. The viewing volume 1 by the object image 1 of the camera 1 is an area where an object obtained from the object image 1 of the camera 1 can exist, and has a weight shape. The visual volume 2 by the object image 2 of the camera 2 is an area where an object obtained from the object image 2 of the camera 2 can exist and has a weight shape. The common area of the viewing volume 1 and the viewing volume 2 is an object existence possible area that is an area where an object can exist when the camera 1 and the camera 2 are used.

  The volume of the object existable region can be calculated by counting the minute regions in the space by a method generally called “visual volume intersection method”. Specifically, for example, literature (Toshikazu Wada, Xiaojun Wu, Shogo Tokai, Takashi Matsuyama, “Homography Based Parallel Volume Intersection: Toward Real-Time Volume Reconstruction Using Active Cameras”, In proceedings of Fifth IEEE International Workshop on Computer Architectures for Machine Perception, 2000.) Spacecarving method and Volume interesection method (VIM) can be used. Note that the visual volume intersection method is not limited to this method, and other existing methods may be used.

  FIG. 7 is an explanatory diagram for explaining the calculation of the sum x of the volumes of the object existence possible area using the visual volume intersection method in a general case where a plurality of objects exist. The specific area shown in FIG. 7 indicates an area where an object in real space such as a room should be counted. The specific area is actually a three-dimensional space, but since it does not lose generality, it is represented in two dimensions in FIG. The specific area may be a space that is partitioned by a wall or the like in practice, or may be a virtual area in a space where the wall is not partitioned. It is also possible not to designate the specific area as a limited area (range).

  Three objects shown in FIG. 7 represent objects to be counted in the specific area. Therefore, in the case shown in FIG. 7, the number of objects is “3”. The object existence possible areas 1 to 8 shown in FIG. 7 are the object existence possibility areas obtained by the visual volume intersection method. In the case illustrated in FIG. 7, the object existence possible areas 2, 4, and 8 are portions that actually include the object. The other object existence possible areas 1, 3, 5, 6, and 7 do not actually include an object, but the object is actually included only from the captured image of the camera 1 and the captured image of the camera 2. It cannot be judged whether or not.

  In the present embodiment, the sum x of the volumes of all the object existence possible regions 1 to 8 calculated using the visual volume intersection method is used for estimating the number of objects. The sum of the volumes of all the object existence possible areas 1 to 8 may be simply the number of minute areas such as voxels in the real space, or may be a scale such as a cubic centimeter, a cubic metric, and a liter. Well, not particularly limited.

  Based on the foreground image of the camera 1, the foreground image of the camera 2, and the camera position / angle parameters of each camera, the object existence possible area calculation unit 32 performs the object in the real space. The sum x of the volumes of the possible regions is calculated using the visual volume intersection method.

  In the next step S112, the object number estimated value calculation unit 34 receives the sum x of the volume of the object existable region received from the object existable region calculation unit 32 and the parameters accumulated in the object number estimation parameter accumulation unit 24. Are used to calculate the estimated number of objects. The method for calculating the estimated value of the number of objects uses regression analysis, which is a commonly used statistical method, in which the sum x of the volume of the object existable region is an explanatory variable and the number of objects is an objective variable.

・・・（１）式

（１）式のパラメータであるｗ_０及びｗ_１の値は、物体数が事前に分かっている学習用画像から作られた物体存在可能領域の体積の和ｘと既知の物体数との多数の組み合わせを用いて、パラメータ学習部１４で回帰分析により予め決定して、物体数推定用パラメータ蓄積部２４に蓄積させておく（詳細後述）。 In regression (single regression) based on the simplest linear equation, if the estimated value of the number of objects is y, the regression equation can be expressed by the following equation (1).

... (1) Formula

The values of w ₀ and w ₁ that are the parameters of the equation (1) are a large number of the sum x of the volume of the object existence possible area made from the learning image whose number of objects is known in advance and the number of known objects. Using the combination, the parameter learning unit 14 determines in advance by regression analysis, and stores it in the object number estimation parameter storage unit 24 (details will be described later).

・・・（２）式

・・・（３）式

次のステップＳ１１４では、計算した物体数ｙを所定の出力先に出力した後、本処理を終了する。なお、物体数ｙを整数値として出力する場合は、回帰分析により算出したｙそのものではなく、下記（４）式で示した物体数ｙを四捨五入した値ｙ’等を出力する。

・・・（４）式

物体数ｙの出力先は、例えば、ディスプレイやタブレット端末等の表示装置であってもよい。また、例えば、出力先は、物体数推定装置１０内の物体数蓄積部（図示省略）等の記憶部等であってもよい。このように出力先は、特に限定されるものではなく、ユーザの所望に応じて予め定めておけばよい。 In equation (1), a linear equation was used as a regression equation. However, as expressed in equation (2) below, when the equation is expanded to n-order equation, or as expressed in equation (3) below, a variable Can be expanded by a predetermined function φ _i (x), the estimated number of objects can be calculated by using multiple regression analysis which is regression analysis using a plurality of explanatory variables.

... (2) formula

... (3) formula

In the next step S114, the calculated number of objects y is output to a predetermined output destination, and then this process is terminated. When the number of objects y is output as an integer value, a value y ′ obtained by rounding off the number y of objects shown in the following equation (4) is output instead of y calculated by regression analysis.

... (4) formula

The output destination of the number of objects y may be a display device such as a display or a tablet terminal, for example. Further, for example, the output destination may be a storage unit such as an object number accumulating unit (not shown) in the object number estimating apparatus 10. Thus, the output destination is not particularly limited, and may be determined in advance according to the user's desire.

  In this embodiment, the object number y that is an estimated value can be output by the object number estimation process in this way.

  Next, determination of parameters used in the regression analysis of the object number estimated value calculation unit 34 of the object number estimation unit 12 will be described.

  First, a learning image is acquired from each of the camera 1 and the camera 2 over a plurality of times in which at least one of the number, size, shape, and position of the objects is different. In the present embodiment, the learning image refers to an image captured in a state where objects of a known number of objects are arranged in a specific area for parameter learning. The acquired learning images of the camera 1 and the camera 2 are associated with the number of objects counted manually to obtain learning data, and a plurality of learning data obtained a plurality of times are stored in a memory (not shown). .

The object existence possible area calculated by the view volume intersection method, the foreground image, and the like vary depending on the state and condition of the object such as the position, size, and shape of the object, and can take various values. Therefore, in the present embodiment, by changing the state or condition of the object, the foreground image is extracted and the object existence possible region is extracted using the learning image captured in the changed state as described later. A plurality of combinations of the volume sum x _tr and the known number of objects are obtained. As a specific example, in the present embodiment, a learning image is acquired at a predetermined timing while an object is randomly moved within a specific region. In the present embodiment, the learning image is acquired at a predetermined timing while the number of objects is changed and the objects are similarly moved randomly. The method of changing the state and condition of the object and the number of learning data to be acquired are not particularly limited as long as a meaningful parameter can be obtained in the regression analysis, and an existing method may be used.

  It should be noted that the change of the state and condition of the object is not performed by the object number estimation device 10 but is performed by the user or the like.

  FIG. 8 shows a flowchart of an example of parameter learning processing in the parameter learning unit 14 of the present embodiment.

  The parameter learning process is executed in advance before the object number estimation unit 12 performs the object number estimation process. The execution timing may be executed, for example, when a camera for estimating the number of objects is newly provided or when the camera settings (shooting conditions including the shooting position) are changed. The execution timing is not particularly limited, and may be executed as desired by the user. In any case, the CPU of the object number estimation apparatus 10 executes the object number estimation program stored in the ROM, whereby the parameter learning process shown in FIG. 8 is executed. Since the parameter learning process includes the same process as the above-described object number estimation process, the same process is described as such and a detailed description thereof is omitted. In the present embodiment, the shooting condition refers to conditions for shooting on the shooting side (camera).

  First, in step S200, the learning foreground extraction unit 40 acquires the learning image of the camera 1 of one piece of learning data from the memory.

  In the next step S <b> 202, the learning foreground extraction unit 40 acquires a background image corresponding to the camera 1 from the background image storage unit 20. As described above, in the background image storage unit 20, background images captured by the camera 1 in a state where no object exists in the specific area are stored in association with the camera 1 in advance.

  In the next step S204, a foreground image is extracted from the learning image based on the learning image captured by the camera 1 and the background image acquired from the background image storage unit 20. The method for extracting the foreground image in the learning foreground extraction unit 40 is the same as the method for extracting the foreground image in the foreground image extraction unit 30 described above. That is, the learning foreground extraction unit 40 extracts the foreground image using the difference between the learning image and the background image.

  In the next step S206, it is determined whether foreground images have been extracted from the learning images for all cameras. If not extracted, for example, in the present embodiment, if only the foreground image of the camera 1 has been extracted, the process returns to step S200, and this process is repeated to extract the foreground image from the learning image of the camera 2. On the other hand, when foreground images are extracted from the learning images of all cameras, the process proceeds to step S208. Instead of repeating the process of extracting the foreground for each camera as described above, the process of each step may be performed simultaneously on the learning images of all the cameras.

  In step S <b> 208, the learning object possible area calculation unit 42 acquires the positions (photographing positions) and angle parameters of the camera 1 and camera 2 from the camera position / angle parameter storage unit 22.

In the next step S210, the learning object existence possible area calculation unit 42 performs an actual operation based on the foreground image of the camera 1 extracted by the learning foreground extraction unit 40, the foreground image of the camera 2, and the camera position / angle parameters of each camera. The sum x _tr of the volume of the object existence possible area in the space is calculated using the visual volume intersection method. The calculation method of the sum x _tr of the volume of the object existence possible area by the visual volume intersection method in the learning object possible area calculation unit 42 is the volume of the object existence possible area by the visual volume intersection method in the object existence possible area calculation unit 32 described above. This is the same as the method of calculating the sum x _tr of.

  In the next step S212, the object number estimation parameter determination unit 44 acquires the number of known objects of the learning data from the memory.

In the next step S214, the object number estimation parameter determination unit 44 stores the acquired number of objects in association with the sum x _tr of the volume of the object existence possible area received from the learning object existence area calculation unit 42. To do.

In the next step 216, it is determined whether or not the processing in steps S200 to S214 has been executed for all learning data. If the size or arrangement of the object depends on the environment of the specific area where the camera is installed, learning data corresponding to the state or condition of the object considering the environment is acquired, and the object existence possible area By acquiring the combination of the volume sum x _tr and the number of known objects, the accuracy of estimating the number of objects by the object number estimation device 10 can be further improved.

In step 216, if there is learning data that has not been subjected to the processing in steps S200 to S214, the process returns to step 200, and the processing in steps S200 to S214 is repeated for the learning data to determine the object existence possible area. The combination of the volume sum x _tr and the known number of objects is acquired and held by the object number estimation parameter determination unit 44.

On the other hand, when the processing of steps S200 to S214 is executed for all the learning data, for example, when the combination of the volume sum x _tr of the object existable regions and the number of known objects is acquired, the process proceeds to step S218. move on.

In step S218, the regression analysis used in the object number estimation value calculation unit 34 is performed from all combinations of the volume sum x _tr of the object _existable regions held by the object number estimation parameter determination unit 44 and the known number of objects. Determine the parameters. Specifically, when the regression analysis represented by the above equation (1) is performed, the parameters refer to the parameters w ₀ and w ₁ . Further, the (2) When performing a regression analysis of the formula is a parameter, the parameter _{w 0,} w 1, refers to · · · _{w n.}

In the next step S220, the object number estimation parameter determination unit 44 stores the determined parameters in the object number estimation parameter storage unit 24, and then ends the present process.
(Second Embodiment)
In the present embodiment, in addition to the object existence possible region, the feature of the two-dimensional image such as the area of the foreground image extracted by the foreground image extraction unit 30 is used as an explanatory variable for regression analysis to determine the number of objects in the specific region. Make an estimate. Since the object number estimation device of the present embodiment includes the same configuration and operation (processing) as the object number estimation device 10 of the first embodiment, the same configuration and operation (processing) are described as such. Detailed description will be omitted.

  First, a schematic configuration of the object number estimation apparatus 10 of the present embodiment will be described. As in the first embodiment, the object number estimation apparatus 10 of the present embodiment uses a computer, and the CPU executes a program stored in the ROM so that the number of objects whose details will be described later is used. An estimation process is executed.

  FIG. 9 is a block diagram illustrating an example of a functional configuration of the object number estimation apparatus 10 according to the present embodiment. As shown in FIG. 9, in the object number estimation device 10 of the present embodiment, the object number estimation unit 12 includes an image area / feature calculation unit 38 as compared to the object number estimation device 10 of the first embodiment. The parameter learning unit 14 includes a learning image area / feature calculation unit 48.

  The image area / feature calculation unit 38 acquires the foreground image of the camera 1 and the foreground image of the camera 2 from the foreground image extraction unit 30, and calculates the feature of each foreground image. The image area / feature calculation unit 38 calculates an image feature from the captured image of the camera 1 and the captured image of the camera 2. The calculated image feature of each captured image is output to the object number estimated value calculation unit 34 for each captured image as an image feature amount representing the image feature.

  The learning image area / feature calculation unit 48 acquires the foreground image of the camera 1 and the foreground image of the camera 2 from the learning foreground extraction unit 40, and calculates the image feature of each foreground image. The learning image area / feature calculation unit 48 calculates image features from the learning image of the camera 1 and the learning image of the camera 2. The calculated image feature of each learning image is output to the object number estimated value calculation unit 34 for each learning image as an image feature amount representing the image feature.

  Next, the object number estimation process executed by the object number estimation apparatus 10 of the present embodiment will be described. FIG. 10 shows a flowchart of an example of the object number estimation process executed by the object number estimation apparatus 10 of the present embodiment.

  Step S300, step S302, step S304, and step S306 of the object number estimation process of the present embodiment are steps S100, S102, and the object number estimation process (see FIG. 5) executed in the first embodiment. It is the same as step S104 and step S106.

  That is, in step S300, the foreground image extraction unit 30 acquires a captured image input from the camera 1. In the next step S302, the foreground image extraction unit 30 acquires a background image corresponding to the camera 1 from the background image storage unit 20. In the next step S304, a foreground image corresponding to the object image is extracted based on the photographed image photographed by the camera 1 and the background image acquired from the background image storage unit 20.

  In the next step S306, it is determined whether foreground images have been extracted from the captured images of all cameras. If not extracted, the process returns to step S300 to repeat this process. On the other hand, if foreground images are extracted from the images captured by all cameras, the process proceeds to step S308. In the present embodiment, in step S306, the process proceeds to step S308 and to step S312.

  Steps S308 and S310 of the object number estimation process of the present embodiment are the same as steps S108 and S110 of the object number estimation process (see FIG. 5) executed in the first embodiment.

  That is, in step S <b> 308, the object existence possible area calculation unit 32 acquires the camera 1 and camera 2 positions (shooting positions) and angle parameters from the camera position / angle parameter storage unit 22. In the next step S310, the object existence possible area calculation unit 32 calculates the real space based on the foreground image of the camera 1 extracted by the foreground image extraction unit 30, the foreground image of the camera 2, and the camera position / angle parameters of each camera. Is calculated using the view volume intersection method, and the sum x of the volumes of the object existable areas is calculated.

  On the other hand, in step S312, the image area / feature calculation unit 38 calculates the areas of the foreground image of the camera 1 and the foreground image of the camera 2 extracted by the foreground image extraction unit 30. In the present embodiment, when a plurality of object images are included in the foreground image, the sum of the areas of all the object images is calculated. In addition, the area of each acquired foreground image is calculated, and the areas of each foreground image are not added. As a method for calculating the area of the foreground image, for example, the number of pixels in the foreground may be counted to calculate the area on the foreground image. The unit may be the number of pixels, or a normalized value.

  In the next step S 314, the image area / feature calculation unit 38 acquires the captured image of the camera 1 and the captured image of the camera 2 from the foreground image extraction unit 30. Note that the captured image of the camera 1 and the captured image of the camera 2 may not be acquired from the foreground image extraction unit 30, but may be directly acquired from the camera 1 and the camera 2 as in the foreground image extraction unit 30. .

  The image area / feature calculation unit 38 calculates an image feature amount for each captured image and outputs the image feature amount to the object number estimated value calculation unit 34. That is, image feature amounts corresponding to the number of captured images (two in the present embodiment) are output. The image feature amount is, for example, the sum of the edge lengths calculated by performing image processing such as edge extraction, but is not limited to this, and any image feature amount representing an image feature There is no particular limitation, and a known method may be used. Note that the image area / feature calculation unit 38 is not limited to the image feature amount related to one image feature (the edge in the above example), and calculates the image feature amount related to a plurality of types of image features, and estimates the plurality of values to the number of objects. You may make it output to the value calculation part 34. FIG. Hereinafter, the image feature amount and the area of the foreground image are referred to as an image feature.

  In the next step S316, the object number estimated value calculation unit 34 calculates the sum x of the volume of the object existable region acquired from the object existable region calculation unit 32 and the image feature amount acquired from the image area / feature calculation unit 38. The estimated number of objects is calculated using the parameters accumulated in the parameter accumulation unit 24 for estimating the number of objects. Also in the present embodiment, the method for calculating the estimated value of the number of objects is a commonly used statistical method using the sum x of the volume of the object existable region and the image feature amount as explanatory variables and the number of objects as an objective variable. A regression analysis is used. In this embodiment, since there are a plurality of explanatory variables, multiple regression analysis is performed.

・・・（５）式

・・・（６）式

（６）式において、ａは、回帰係数を並べた縦ベクトルで、ベクトル「ｘ^〜」と同じ次元数である。 The sum x of the object existence possible area and the image feature amount are expressed by a vector “x”. When a new vector represented by the following equation (5) with “1” added to the first element of the vector “x” is introduced, the multiple regression analysis for estimating the number of objects y represents the regression equation (6 ) Expression.

... (5) formula

... (6) formula

In (6), a is the vertical vector obtained by arranging regression coefficients, the same number of dimensions as vector "x ^~."

・・・（７）式

・・・（８）式

次のステップＳ３１８は、計算した物体数ｙを、第１の実施の形態と同様に、所定の出力先に出力した後、本処理を終了する。 Further, similarly to the above (3), the introduction of the function represented by the following equation (7) for outputting a scalar value from the vector inputs may represent a vector "x ^~" in the following equation (8).

... (7) formula

... (8) formula

In the next step S318, the calculated number of objects y is output to a predetermined output destination in the same manner as in the first embodiment, and then this process is terminated.

  In this embodiment, the object number y that is an estimated value can be output by the object number estimation process in this way.

  Next, determination of parameters used in the regression analysis of the object number estimation value calculation unit 34 of the object number estimation unit 12 in the present embodiment will be described. FIG. 11 shows a flowchart of an example of parameter learning processing in the parameter learning unit 14 of the present embodiment.

  Step S400, step S402, step S404, and step S406 of the parameter learning process of the present embodiment are steps S200, S202, and S204 of the parameter learning process (see FIG. 8) executed in the first embodiment. , And step S206.

  That is, in step S400, the learning foreground extraction unit 40 acquires the learning image of the camera 1 of one learning data from the memory. In the next step S <b> 402, the learning foreground extraction unit 40 acquires a background image corresponding to the camera 1 from the background image storage unit 20. In the next step S404, the foreground image corresponding to the object image is extracted based on the learning image captured by the camera 1 and the background image acquired from the background image storage unit 20.

  In the next step S406, it is determined whether or not foreground images have been extracted from the images captured by all cameras. If not extracted, the process returns to step S400 to repeat this process. On the other hand, if foreground images have been extracted from the captured images of all cameras, the process proceeds to step S408. In the present embodiment, similarly to the object number estimation process, in step S406, the process proceeds to step S408 and to step S412.

  Step S408 and step S410 of the parameter learning process of the present embodiment are the same as step S208 and step S210 of the parameter learning process (see FIG. 8) executed in the first embodiment.

That is, in step S <b> 408, the learning object possible area calculation unit 42 acquires the positions (photographing positions) and angle parameters of the cameras 1 and 2 from the camera position / angle parameter storage unit 22. In the next step S410, the learning object existence possible area calculation unit 42 performs an actual operation based on the foreground image of the camera 1 extracted by the learning foreground extraction unit 40, the foreground image of the camera 2, and the camera position / angle parameters of each camera. The sum x _tr of the volume of the object existence possible area in the space is calculated using the visual volume intersection method.

  On the other hand, in step S 412, the learning image area / feature calculation unit 48 calculates the areas of the foreground image of the camera 1 and the foreground image of the camera 2 extracted by the foreground image extraction unit 30. In the present embodiment, when a plurality of object images are included in the foreground image, the sum of the areas of all the object images is calculated. The method for calculating the area of the foreground image by the learning image area / feature calculation unit 48 may be the same as that of the image area / feature calculation unit 38 described above.

  In the next step S 414, the learning image area / feature calculation unit 48 acquires the captured image of the camera 1 and the captured image of the camera 2 from the learning foreground extraction unit 40.

  Further, the learning image area / feature calculation unit 48 outputs the image feature amount calculated for the image feature amount to the object number estimation parameter determination unit 44 for each learning image. The image feature amount and its calculation method may be the same as those of the image area / feature calculation unit 38 described above.

  The next step S416, step S418, and step S420 are the same as step S212, step S214, and step S216 of the parameter learning process (see FIG. 8) executed in the first embodiment.

That is, in step S416, the object number estimation parameter determination unit 44 acquires the known number of objects of the learning data from the memory. In the next step S418, the object number estimation parameter determination unit 44 stores the acquired number of objects in association with the sum x _tr of the volume of the object existence possible area received from the learning object existence area calculation unit 42. To do. In the next step S420, the object number estimation parameter determination unit 44 determines whether or not the processing in steps S400 to S418 has been executed for all learning data. If there is learning data that has not been subjected to the processing of steps S400 to S418, the process returns to step 400, and the processing of steps S400 to S418 is repeated for the learning data, and the volume x of the object existence possible area x The combination of _tr and the known number of objects is acquired and held by the object number estimation parameter determination unit 44. On the other hand, when the processing of steps S400 to S418 has been executed for all learning data, the process proceeds to step S422.

In step S422, the object number estimation value calculation unit 34 is obtained from all combinations of the volume sum x _tr of the object existence possible area held by the object number estimation parameter determination unit 44, the image feature amount, and the number of known objects. Determine the parameters of the regression analysis used in. Specifically, the vector a composed of each parameter of the multiple regression analysis described in the object number estimation process of the present embodiment is determined.

  When there are a large number of explanatory variables in the multiple regression analysis, if there are combinations of variables having a high correlation among them, the operation may become unstable. In order to avoid this, for example, generally known techniques such as those described in the literature (Trevor Hastie, Robert Tibshirani, Jerome Friedman, “The Elements of Statistical Learning,” Springer, 2001.) are used. It may be used. Also in the first embodiment, such a technique is preferably used when performing multiple regression analysis.

  Also, instead of introducing the function φ (x) in general regression including nonlinearity, for example, literature (Nello Cristianini, John Shawe-Taylor, Takeshi Ohkita (translation), “Introduction to Support Vector Machine”, Kyoritsu Publishing , 2005.), support vector regression (SVR) may be used.

  In the next step S420, the object number estimation parameter determination unit 44 stores the determined parameters in the object number estimation parameter storage unit 24, and then ends this process.

As described above, according to each of the above embodiments, the object number estimation device 10 includes the object number estimation unit 12, the parameter learning unit 14, the background image storage unit 20, the camera position / angle parameter storage unit 22, and the object. A number estimation parameter storage unit 24 is provided. The foreground image extraction unit 30 extracts the image corresponding to each captured image from the captured images captured by the cameras 1 and 2 and the background image corresponding to each camera stored in the background image storage unit 20. The object existence possible area calculation unit 32 calculates the volume of the object existence possibility area by the visual volume intersection method based on the position and angle parameters of each camera acquired from the camera position / angle parameter accumulation unit 22 and each foreground image. Calculate the sum x. The object number estimated value calculation unit 34 calculates and outputs an estimated value of the number of objects using a regression analysis parameter in which the sum x of the volume of the object existable region is an explanatory variable and the number of objects y is an objective variable. The parameters for regression analysis used by the object number estimation value calculation unit 34 are determined in advance by the object number estimation parameter determination unit 44 of the parameter learning unit 14 and stored in the object number estimation parameter storage unit 24. Based on the learning data, the object number estimation parameter determination unit 44 includes the object presence in the learning image calculated in the same manner as described above by the learning foreground extraction unit 40 and the learning object existence possible region calculation unit 42 of the parameter learning unit 14. The regression analysis parameters are determined with the sum of the volumes of the possible regions x _tr as the explanatory variable and the known number of objects y _tr as the objective variable.

  As described above, the object number estimation apparatus 10 estimates the number of objects in the specific area using the captured images captured at different capturing positions. Therefore, the captured image captured at one capturing position is the previous one. It is possible to consider the position of the object behind that is hidden by the object.

Further, since regression analysis which is statistical analysis by the parameter learning unit 14 is used, it is possible to deal with various object states and conditions. The sum x, _xtr of the volume of the object _existable area calculated by the object existable area calculating unit 32 and the learning object possible area calculating unit 42 is the individual difference of the object, the difference in the visual volume depending on the position of the object, and the object Even if the number of objects is the same due to the arrangement, various values can be taken. However, since regression analysis, which is statistical analysis, is used in the present invention, such an individual difference in an object, a difference in visual volume depending on the position of the object, an arrangement of the object, and the like can be determined in a specific region. The estimated number of objects can be accurately calculated and output.

  Therefore, the number of objects existing in the specific area can be accurately estimated based on the captured images captured at a plurality of capturing positions according to the present invention.

  In the second embodiment, since the image features calculated by the image area / feature calculation unit 38 and the learning image area / feature calculation unit 48 are also used as explanatory variables for regression analysis, the estimation accuracy of the number of objects is increased. Can be better.

  In each of the above embodiments, the case where two cameras are used, that is, the case where there are two shooting positions has been described, but the present invention is not limited to this. The number of shooting positions is not particularly limited as long as it is two or more, and the number of shooting positions may be a number according to a specific area, a user's desire, or the like. In addition, you may image | photograph from several imaging | photography positions, when one camera moves. In this case, for example, the shooting position and the background image are associated with each other and stored in the background image storage unit 20, and the camera position and angle parameter of each shooting position are stored in the camera position / angle parameter storage unit 22. It is preferable that information indicating the shooting position is added to the captured image and input.

  In each of the upper embodiments, the case where the captured image and the learning image are still images has been described, but a moving image may be used. When using a moving image, the number of objects can be estimated in the same manner as in each of the above embodiments by using an image of each frame corresponding to a still image.

  In each of the above embodiments, the case where the object whose number of objects is estimated is a person. However, the present invention is not limited to this, and an object other than a person (for example, a vehicle or the like) may be used.

  In addition, the configurations and operations of the object number estimation device 10, the object number estimation unit 12, and the parameter learning unit 14 described in the above embodiments, and the object number estimation process and the parameter learning process are examples. Needless to say, it can be changed according to the situation without departing from the spirit of the present invention.

DESCRIPTION OF SYMBOLS 10 Object number estimation apparatus 12 Object number estimation part 14 Parameter learning part 30 Foreground image extraction part 32 Object existence possible area | region calculation part 34 Object number estimated value calculation part 38 Image area and feature calculation part

Claims (6)

Difference between a plurality of background images obtained by photographing a specific area from a plurality of photographing positions in the absence of an object in advance and a plurality of photographed images obtained by photographing the specific area from the plurality of photographing positions by the photographing foreground image extracting means Extracting a foreground image that is an image representing an object from each of the plurality of captured images based on
Based on the imaging conditions including the plurality of imaging positions and the plurality of foreground images extracted by the imaging foreground image extraction unit by the object possible region calculation unit, Calculating the volume of each possible region of the object at
The number of objects was previously determined by the number-of-objects estimation unit based on a plurality of learning images obtained by photographing the plurality of background images and the specific region where the number of objects of a known number of objects existed at the plurality of photographing positions. Existence of each object calculated by the object existence area calculation means using a regression analysis parameter with the sum of each volume of the object existence area as an explanatory variable and the number of known objects as an objective variable Calculating and outputting an estimate of the number of objects in the specific region based on the sum of the volumes of possible regions; and
An object number estimation method comprising: Based on the difference between the plurality of background images and the plurality of learning images obtained by photographing the specific area where the objects of the known number of objects exist at the plurality of photographing positions by the learning foreground image extracting unit. Extracting a learning foreground image, which is an image representing an object, for each of the learning images;
Based on the imaging conditions and the plurality of learning foreground images, the learning object existence possible area calculation means calculates the volume of each of the object existence possible areas in the specific area obtained using the visual volume intersection method. And steps to
Determining, by parameter determining means, parameters for regression analysis with the sum of the volumes of the possible areas of the objects calculated by the learning object possible area calculating means as an explanatory variable and the number of known objects as an objective variable; ,
The object number estimating means performs regression analysis by calculating and outputting an estimated value of the number of objects in the specific region using the parameter determined by the parameter determining means.
The object number estimation method according to claim 1. A step of calculating a predetermined image feature for the learning image and the learning foreground image by a learning image feature calculating means;
Calculating the predetermined image feature for the photographed image and the foreground image by an image feature calculating means,
The parameter determination means is calculated for the learning image and the learning foreground image by the sum of the volumes of the object existence possible areas calculated by the learning object existence possibility area calculation means and the learning image feature calculation means. Determining a regression analysis parameter with the predetermined image feature as an explanatory variable and the number of known objects as an objective variable,
The object number estimating means includes a sum of volumes of the object existable areas calculated by the object existable area calculating means and a predetermined value calculated for the photographed image and the foreground image by the image feature calculating means. Calculate and output an estimate of the number of objects based on image features;
The object number estimation method according to claim 1 or 2. The predetermined image feature calculated for the learning image and the learning foreground image includes an area of each of the plurality of learning foreground images and a predetermined image feature amount of each of the plurality of learning images.
The predetermined image feature calculated for the captured image and the foreground image includes an area of each of the plurality of foreground images and the predetermined image feature amount of each of the plurality of captured images. Object number estimation method. Based on the difference between a plurality of background images obtained by photographing a specific region from a plurality of photographing positions in the absence of an object in advance and a plurality of photographed images obtained by photographing the specific region from the plurality of photographing positions. Shooting foreground image extracting means for extracting a foreground image that is an image representing an object from each of the shot images;
A region where an object can exist in the specific region obtained by using a visual volume intersection method based on the shooting conditions including the plurality of shooting positions and the plurality of foreground images extracted by the shooting foreground image extracting unit. An object existence possible area calculating means for calculating each volume of
The object existence possible area obtained in advance based on the plurality of background images and the plurality of learning images obtained by photographing the specific area where the objects of a known number of objects exist at the plurality of the photographing positions. Using the regression analysis parameters with the sum of each of the volumes as explanatory variables and the number of known objects as the objective variable, An object number estimating means for calculating and outputting an estimated value of the number of objects in the specific region based on the sum;
An object number estimation device comprising:   An object number estimation program for causing a computer to execute each step of the object estimation method according to any one of claims 1 to 4.

Images