JP6920949B2 - Object distribution estimator - Google Patents

Description

The present invention relates to an object distribution estimation device that estimates the distribution of an object from a photographed image in which a space in which a predetermined object such as a person can exist is photographed. The present invention relates to an object distribution estimation device that estimates a distribution.

In spaces where congestion may occur, such as venues where events such as marathons and parades are held, it is necessary to take measures such as allocating a large number of security guards in areas where congestion is occurring in order to prevent accidents. Therefore, it is expected that the monitoring efficiency will be improved by arranging surveillance cameras at various places in the venue, estimating the distribution status of people from the captured images, and displaying the distribution status.

As one of the methods for detecting a person present at an event venue from a photographed image, there is a method of searching for the photographed image by a recognizer that has learned the characteristics of the person's image in advance.

For example, in the object detection device described in Patent Document 1 below, an input image is used as a recognizer by using a discriminator trained in advance using a large number of "human" image data and "non-human" image data. It is described to detect a person.

Further, in the crowd analysis device described in Patent Document 2 below, the person density is estimated by using a discriminator machine-learned for each person density using a learning image in which a crowd having a person density exceeding the lower limit of density is captured in advance. It is stated that by doing so, the occurrence of the crowd is determined.

Hereinafter, a recognizer that recognizes each person is referred to as a single classifier, and a recognizer that estimates the person density is referred to as a density estimator.

Japanese Unexamined Patent Publication No. 2011-186633 Japanese Unexamined Patent Publication No. 2017-068598

By the way, since the event venue is generally vast, it is required to widen the field of view of each surveillance camera and reduce the number of installed cameras from the viewpoint of installation and operation costs.

However, the single classifier and the density estimator differ in the range of resolutions capable of highly accurate recognition depending on the characteristics of each trained image. On the other hand, in a captured image taken in a wide field of view, the resolution of the image of a person is significantly reduced according to the distance from the camera to the person.

Therefore, there is a problem that it is difficult to provide highly accurate information when the recognizer is used without considering the characteristics of the captured image taken in a wide field of view.

That is, while the single classifier can recognize the position of each person, the density estimator cannot recognize the position of each person. In general, when the resolution of the image to be recognized (a part of the captured image) is lower than that of the learning image, the recognition accuracy decreases, and the learning of the single classifier uses the learning image occupied by the image of a single person. On the other hand, since the learning of the density estimator is performed using the learning image of the field where images of multiple people can be captured, the recognition accuracy decrease according to the distance from the camera to the person is more than the density estimator. However, a single classifier is more likely to occur.

Therefore, it is possible to provide detailed information by recognizing the distribution of people in the area near the camera in the captured image with a single classifier, but if the distribution of people in a distant area is also recognized by the single device, the accuracy is high. It will provide low information. On the other hand, it is possible to provide more accurate information by recognizing the distribution of people in the distant area with a density estimator, but if the density estimator tries to recognize even the distribution of people in the area near the camera, detailed information can be obtained. It will fail to provide.

The present invention has been made in view of the above problems, and provides an object distribution estimation device capable of estimating highly accurate information on the distribution of an object in a wide range from a photographed image of a space where congestion may occur in a wide field of view. The purpose is.

(1) The object distribution estimation device according to the present invention has an image acquisition means for acquiring a photographed image of a space in which a predetermined object photographed by a photographing unit may exist, and each has a different number of photographed objects. It is a recognition means that recognizes the number of the objects photographed at an arbitrary position in the photographed image by using a recognizer that has learned the characteristics of each of the plurality of types of learning images in advance, and the ranges of the number of recognizable objects are mutual. A plurality of recognition means provided for each range of the number of objects using a plurality of different recognizers, a position in the captured image, and the number of objects at the position among the plurality of recognition means in advance. From the number of objects acquired in the captured image by the storage means that stores the accuracy guarantee recognition means that can be recognized with the accuracy equal to or higher than the predetermined lower limit value and the accuracy guarantee recognition means that is stored in the storage means. , A distribution estimation means for generating distribution information of the object in the space.

(2) In the object distribution estimation device according to (1) above, the lower limit of the resolution of the image of the object that the recognition means can recognize with an accuracy equal to or higher than the lower limit is defined as the limit resolution of the recognition means. As the storage means, among the plurality of recognition means, the one in which the resolution of the image of the object at the position is equal to or higher than the limit resolution with respect to the position in the captured image is used as the accuracy guarantee recognition means. It can be associated and stored.

(3) In the object distribution estimation device according to (2) above, the resolution is represented by the number of pixels of the image of the object, and the number of pixels corresponding to the limit resolution is set as the limit number of pixels and exists in the space. A projection means for generating a projected image by projecting a model imitating the object onto the photographing surface of the photographing unit, the number of pixels of the projected image generated at the position with respect to the position in the photographed image, and the above. The configuration may further include a selection means for selecting the accuracy guarantee recognition means associated with the position by comparing the limit number of pixels of each of the plurality of recognition means.

(4) In the object distribution estimation device according to (2) above, the resolution is estimated from the distance from the photographing unit to the object in the space, and the distance corresponding to the limit resolution is set as the limit distance and the storage is performed. The means associates the plurality of recognition means whose distance with respect to the object photographed at the position is equal to or less than the limit distance as the accuracy guarantee recognition means with respect to the position in the captured image. It can be a memorized configuration.

(5) In the object distribution estimation device according to (4) above, a projection means for obtaining an object virtual position by back-projecting a position in the captured image onto a horizontal plane having a height corresponding to the object in the space, and the photographing. The accuracy guarantee recognition associated with the position by comparing the distance between the virtual position of the object corresponding to the position and the photographing unit with respect to the position in the image and the limit distance of each of the plurality of recognition means. The configuration may further include a selection means for selecting the means.

(6) Another object distribution estimation device according to the present invention includes an image acquisition means for acquiring a photographed image of a space in which a predetermined object photographed by a photographing unit may exist, and the object distribution estimation device according to the present invention at the density at each predetermined density. A density estimation means that recognizes the density of the object photographed at an arbitrary position in the photographed image by using a density estimator in which the characteristics of each density image obtained by photographing the space in which the object exists are learned in advance, and a single unit. A unit identification means for recognizing the presence or absence of the object photographed at an arbitrary position in the photographed image by using a unit classifier in which the characteristics of the unit image in which the object is photographed are learned in advance, and a position in the photographed image. And a storage means that stores the density or presence / absence of the object at the position in association with the density estimation means and the accuracy guarantee recognition means that can recognize the presence / absence of the object with an accuracy equal to or higher than a predetermined lower limit value. A distribution estimation means that generates distribution information of the object in the space from the density or presence / absence of the object acquired in the captured image by the accuracy guarantee recognition means stored in the storage means.

According to the present invention, highly accurate information on the distribution of objects can be estimated in a wide range from a photographed image of a space where congestion may occur in a wide field of view.

It is a block diagram which shows the schematic structure of the object distribution estimation apparatus which concerns on embodiment of this invention. It is a functional block diagram when the object distribution estimation apparatus which concerns on embodiment of this invention performs distribution estimation processing based on photographed images. It is a functional block diagram when the object distribution estimation apparatus which concerns on embodiment of this invention performs a method map creation process. It is a figure which represented typically the information of the simple substance classifier stored in the simple substance classifier storage means. It is a schematic diagram which shows an example of the photographed image and the method map corresponding thereto. It is a schematic diagram which shows an example of the distribution image corresponding to the photographed image of FIG. It is a schematic diagram which shows an example of the method map corresponding to the photographed image of FIG. It is a schematic flow chart of the operation of the object distribution estimation apparatus which concerns on embodiment of this invention. It is a schematic flow chart of the method map creation process. It is a schematic flow chart of the distribution image generation processing.

Hereinafter, the object distribution estimation device 1 according to the embodiment of the present invention (hereinafter referred to as the embodiment) will be described with reference to the drawings.

[Structure of object distribution estimation device 1]
FIG. 1 is a block diagram showing a schematic configuration of the object distribution estimation device 1. The object distribution estimation device 1 includes a photographing unit 2, a communication unit 3, a storage unit 4, an image processing unit 5, an operation unit 6, a display control unit 7, and a display unit 8.

The photographing unit 2 is a surveillance camera, and is connected to the image processing unit 5 via the communication unit 3, and is a photographing means that photographs a predetermined space in which a predetermined object can be congested at a predetermined time interval and outputs a captured image. Is. Hereinafter, the portion photographed by the photographing unit 2 is referred to as a target space.

For example, the photographing unit 2 is installed on a pole installed at the event venue with a field of view that gives a bird's-eye view of the target space. The field of view may be fixed or may be changed according to an instruction from the outside via the communication unit 3. Further, for example, the photographing unit 2 photographs the target space with a frame period of 1/5 second to generate a color image. A monochrome image may be generated instead of the color image.

The communication unit 3 is a communication circuit, one end of which is connected to the image processing unit 5, and the other end of the communication unit 2 and the display control unit 7 via a communication network such as a coaxial cable, LAN (Local Area Network), or the Internet. Is connected with. The communication unit 3 acquires a photographed image from the photographing unit 2 and inputs it to the image processing unit 5, and outputs the information input from the image processing unit 5 to the display control unit 7.

The storage unit 4 is a memory device such as a ROM (Read Only Memory) and a RAM (Random Access Memory), and stores various programs and various data. The storage unit 4 is connected to the image processing unit 5 and inputs / outputs these information to and from the image processing unit 5.

The image processing unit 5 is composed of arithmetic units such as a CPU (Central Processing Unit), a DSP (Digital Signal Processor), and an MCU (Micro Control Unit). The image processing unit 5 is connected to the storage unit 4 and operates as various processing means / control means by reading and executing a program from the storage unit 4, storing various data in the storage unit 4, and also from the storage unit 4. read out. Further, the image processing unit 5 is also connected to the photographing unit 2 and the display control unit 7 via the communication unit 3, and the person being photographed by analyzing the photographed image acquired from the photographing unit 2 via the communication unit 3. Is estimated, and distribution information such as a distribution image in which the estimation result is described is output to the display control unit 7 via the communication unit 3.

The operation unit 6 is an input device for the display control unit 7, and is composed of a keyboard, a mouse, and the like. The operation unit 6 is connected to the display control unit 7, receives an instruction operation by the observer, and outputs the instruction operation to the display control unit 7.

The display control unit 7 is an interface circuit with a communication network to which a storage unit (not shown) composed of a PC (Personal Computer) or the like and a memory device such as a ROM or RAM and a communication unit 3 are connected. It includes a communication unit (not shown) and a control unit (not shown) composed of arithmetic units such as a CPU, MCU, and IC. The display control unit 7 is connected to the communication unit 3 via the communication network, and is also connected to the operation unit 6 and the display unit 8. The display control unit 7 receives the information from the image processing unit 5 from the communication unit 3 and stores it, and at the same time, the operation instruction by the observer is input from the operation unit 6 and displays the information corresponding to the operation instruction among the stored information. Output to unit 8.

The display unit 8 is a display device such as a liquid crystal display or a CRT (Cathode Ray Tube) display, and is connected to the image processing unit 5 via the communication unit 3 and the display control unit 7, and the information generated by the image processing unit 5 is generated. Is a display means for displaying. The observer visually recognizes the displayed information, determines the occurrence of congestion, etc., and takes measures such as changing the staffing as necessary.

In this embodiment, the object distribution estimation device 1 in which the number of the photographing unit 2 and the image processing unit 5 is 1: 1 is illustrated, but in another embodiment, the photographing unit 2 and the image processing unit 5 The number can be many-to-one or many-to-many.

[Function of object distribution estimation device 1]
2 and 3 are functional block diagrams showing the functions of the object distribution estimation device 1. Of these, FIG. 2 is a functional block diagram when performing distribution estimation processing based on captured images. Further, FIG. 3 is a functional block diagram when the method map creation process is performed prior to the distribution estimation process.

(Distribution estimation processing)
Among the functions of the object distribution estimation device 1, first, the distribution estimation process will be described. When performing the distribution estimation process (FIG. 2), the communication unit 3 functions as an image acquisition means 30, a distribution information output means 31, and the like, and the storage unit 4 has a density estimator storage means 40, a single classifier storage means 41, and a method map. It functions as a storage means 42, an object model storage means 43 (not shown in FIG. 2), a camera parameter storage means 44 (not shown in FIG. 2), and the like, and the image processing unit 5 has a density estimation means 50 and a single identification means 51. , Functions as a distribution estimation means 52 and the like.

Of these, each of the density estimation means 50 and the single identification means 51 uses a recognizer that has learned in advance the characteristics of each of a plurality of types of learning images in which the number of captured objects is different, and at an arbitrary position in the captured image. It is a recognition means for recognizing the number of objects photographed in the above, and the density estimation means 50 and the single identification means 51 use a plurality of recognizers having different ranges of the number of recognizable objects from each other. It forms a plurality of recognition means provided for each. For example, the density estimating means 50, the density of 0 people / ^{m 2,} 0 people / two / ^{m 2} or less of a density higher than ^{m 2,} higher than two / ^{m 2} Quadruple / ^{m 2} or less density and four Using a recognizer trained using a learning image with a density higher than / m ^{2, the} area inside the window is set to an arbitrary position in the captured image as a window, and ^{the density inside the window is 0 people / m 2} and 0 people / m. two / ^{m 2} or less of a density higher than ^2, recognizes which of higher than two / ^m higher than ² Quadruple / ^{m 2} or less of the density and four / ^{m 2} density. Further, for example, the single identification means 51 uses a recognizer learned using the learning images of 0 people and 1 person, and recognizes whether the inside of the window is 0 people or 1 person. That is, the density estimation means 50 recognizes the density of objects in the window as the number of objects, and the lower limit of the range is 0 people / m ² , while the upper limit reaches a value exceeding ^{4 people / m 2.} On the other hand, the range of the number of objects that the single identification means 51 can recognize in the window is 0 and 1.

The image acquisition means 30 sequentially acquires captured images from the photographing unit 2 which is a photographing means, and sequentially outputs the acquired captured images to the density estimation means 50 and the unit identification means 51.

The density estimator storage means 40 is an estimated density calculation function that learns the image features of each density image obtained by photographing a space in which an object (person) exists at the density at a predetermined density, and obtains the feature amount of the image. When input, the estimated value (estimated density) of the density of the object captured in the image is calculated, and the information of the estimator (density estimator) that outputs the calculated estimated density is stored in advance. That is, parameters such as the coefficient of the estimated density calculation function are stored in advance as information of the density estimator.

The density estimation means 50 extracts the feature amount for density estimation (feature amount for estimation) from various parts of the captured image input from the image acquisition means 30, and reads out the density estimator from the density estimator storage means 40 to extract the feature amount. The distribution of the estimated density (density distribution) is estimated by inputting each of the estimated feature quantities to the density estimator, and the estimated density distribution is output to the distribution estimation means 52. Preferably, the density estimation means 50 also outputs the density distribution to the unit identification means 51.

The processing of density estimation and the density estimator will be specifically described.

The density estimation means 50 sets a window (estimation extraction window) at the position of each pixel of the captured image, and extracts the estimation feature amount from the captured image in each estimation extraction window. The estimation feature quantity is a GLCM (Gray Level Co-occurrence Matrix) feature.

It is desirable that the areas in the target space photographed by each estimation window have the same size. Therefore, preferably, the density estimation means 50 performs processing in consideration of this point. In this process, the density estimation means 50 uses the camera parameter storage means 44 (not shown in FIG. 2).

The camera parameter storage means 44 stores the camera parameters of the photographing unit 2. The camera parameters are information including external parameters such as the installation position and shooting direction of the photographing unit 2 in the real space, the focal length of the photographing unit 2, the angle of view, lens distortion and other lens characteristics, and internal parameters such as the number of pixels of the imaging element. be.

The density estimation means 50 reads out the camera parameters of the photographing unit 2 from the camera parameter storage means 44, and the area in the target space photographed by any pixel of the photographed image becomes the same size by homography conversion using the camera parameters. After transforming the captured image in this way, the feature amount for estimation is extracted.

The density estimator can be realized by a discriminator that discriminates a multi-class image, and can be a discriminant function learned by a multi-class SVM (Support Vector Machine) method.

The density is, for example, a "background" class with no people, a "low density" class ^{higher than 0 people / m 2} and 2 people / m ^{2 or less, higher than 2 people / m 2} and 4 people / m ² or less. It can be defined as 4 classes of "medium density" class and ^{"high density" class higher than 4 people / m 2.}

The estimated density is a value given in advance to each class and is a value output as a result of distribution estimation. In this embodiment, the values corresponding to each class are described as "background", "low density", "medium density", and "high density".

That is, the density estimator learns by applying the multi-class SVM method to the features of a large number of density images belonging to each of the "background" class, "low density" class, "medium density" class, and "high density" class. It is an identification function for distinguishing the image of each class from other classes. The parameters of the discriminant function derived by this learning are stored as a density estimator. The feature amount of the density image is the same as the feature amount for estimation, and is a GLCM feature.

The density estimation means 50 acquires the estimated density, which is the output value, by inputting each of the estimation feature quantities extracted corresponding to each pixel into the density estimator. When the captured image is deformed to extract the feature amount for estimation, the density estimation means 50 transforms the density distribution into the shape of the original captured image by homography conversion using camera parameters.

The set of estimated densities for each pixel of the captured image obtained in this way is the density distribution.

From the density distribution output by the density estimation means 50, the density of people in various parts of the photographed image can be known, but from the density distribution, the position of each person cannot be known. On the other hand, the single identification means 51 is a means for obtaining more detailed information (position of a person) by identifying the presence or absence of a person at each position of the captured image.

The single classifier storage means 41 stores the information of the classifier (single classifier) that has learned the image features of a single person (object).

FIG. 4 is a diagram schematically showing the information of the single-unit classifier stored in advance by the single-unit classifier storage means 41.

When the feature amount of the image is input, the single classifier calculates and outputs an evaluation value (discrimination score) indicating the plausibility that the image is an image (single image) taken by a single person. It is represented by parameters such as the coefficient of the evaluation value calculation function and the threshold value applied to the identification score.

The single classifier can be a classifier learned by applying the linear SVM method to the features of a learning image consisting of a large number of single images and a large number of unmanned images in which only a person is shown.

When linear SVM is used as the learning algorithm, the coefficient of the evaluation value calculation function is a weight vector. This weight vector is a weight for each element of the feature amount, and the value of the inner product of the feature amount of the input image and the weight vector represents the discrimination score. In learning, if the inner product of the weight vector and the feature amount is greater than 0, it is identified as a person, and if it is 0 or less, it is identified as a non-human. Therefore, the threshold value for identifying whether or not the input image is a single image is 0 in principle, and the threshold value can usually be set to 0. However, the threshold value may be set to a value smaller than 0 in order to reduce the error of identifying the single image as a non-single image.

The feature amount of the learning image is a HOG (Histograms of Oriented Gradients) feature amount.

The single-unit classifier stored in the single-unit classifier storage means 41 is a classifier that learns the image features of a small part of the parts constituting a single object as the density increases. The single-unit classifier storage means 41 is independently associated with the whole-body classifier 100, which is a single-unit classifier that has learned the image features of the whole body of a single person in association with the value representing the low-density class, and the value representing the medium-density class. Upper body classifier 101, which is a single classifier that learned the image features of the upper 2/3 of a person, and a single classifier that learned the image features of the upper 1/3 of a single person in association with a value representing a high-density class. The head proximity classifier 102 is stored.

The whole body classifier 100 is a single body classifier learned by using a single image of the whole body of a single person, and the upper body classifier 101 is a single image of the upper two-thirds of a single person (whole body of a person). Is a single classifier learned using an image obtained by cutting out the upper 2/3 of the single image taken by the head, and the head proximity classifier 102 is a single image (such as an image obtained by cutting out the upper 1/3 of a single person). It is a single classifier learned by using an image obtained by cutting out the upper 1/3 of a single image of the whole human body.

The single identification means 51 extracts a feature amount for single identification (feature amount for identification) from each position of the captured image input from the image acquisition means 30, and reads out the single identification device from the single identification device storage means 41. The presence or absence of an object is identified by inputting each of the extracted identification feature amounts to the single classifier, and the presence or absence of the object at each position is output to the distribution estimation means 52.

Specifically, the single identification means 51 first sets a plurality of candidate positions in the captured image at predetermined intervals. The predetermined interval is one pixel, and the single identification means 51 sequentially sets the position of each pixel of the captured image as a candidate position. The candidate position represents the center of gravity of the human head.

Further, the unit identification means 51 sets a window defined in the shape of the upper 1/3 of a single person at each candidate position, and refers to the density distribution input from the density estimation means 50 to estimate the inside of the window. Aggregate the density. Then, the unit identification means 51 determines the highest estimated density at each candidate position as the density of the candidate position.

Further, the single identification means 51 sets an identification extraction window according to the density of the candidate position at each candidate position, and extracts the identification feature amount from the captured image in the identification extraction window. The identification extraction window has the shape of a single image used for learning the single classifier according to each density, and is a window having a size enlarged / reduced by a plurality of predetermined magnifications. That is, the identification extraction window is a window defined to the shape of the whole body of a single person if the density of the candidate positions is low, and is defined to the shape of the upper two-thirds of the single person if the density is medium. It is a window that is shaped like the upper 1/3 of a single person if it has a high density.

Further, the single-unit identification means 51 reads out a single-unit classifier corresponding to the density of the candidate position for each candidate position from the single-unit classifier storage means 41. That is, the single unit identification means 51 reads out the whole body classifier if the density of the candidate positions is low, reads out the upper body classifier if the density is medium, and reads out the head area classifier if the density is high. Then, the single unit identification means 51 inputs the identification feature amount extracted from the candidate position into the read single unit classifier for each candidate position, and acquires the identification score which is the output value as the evaluation value of the candidate position. ..

Then, the unit identification means 51 refers to the density and the evaluation value for each candidate position, and the candidate position where the evaluation value satisfying the predetermined standard is calculated is the position where the object exists, and the other candidate positions are the objects. Determine the position not to be.

Specifically, the single identification means 51 extracts candidate positions whose identification score is equal to or higher than the corresponding threshold value, and selects a plurality of candidate positions having the same density and close to each other among the extracted candidate positions. Summarize them together and determine the summarized candidate positions as the positions where people exist.

This process of summarizing the candidate positions is performed in order to deal with the fact that a high discrimination score is calculated for the same person not only at the position where the person is actually photographed but also in the vicinity thereof. Specifically, for example, the single identification means 51 sets the candidate positions for which the identification score equal to or higher than the threshold value is calculated to the attention positions in descending order of the identification score for each density, and the candidates whose identification score is lower than the attention position. Set the position to the comparison position. Then, the single identification means 51 deletes the information of the comparison position in which the overlap between the identification extraction window set at the comparison position and the identification extraction window set at the attention position is larger than a predetermined ratio among the comparison positions. By doing so, multiple candidate positions are combined into one.

Then, the single unit identification means 51 outputs information on the presence or absence of an object for each candidate position to the distribution estimation means 52.

The method map storage means 42 sets the position in the captured image and the number of objects photographed in the window when the window is set at the position among the plurality of recognition means, which is equal to or higher than a predetermined lower limit value. The method map associated with the accuracy guarantee recognition means that can be recognized with the accuracy of is stored.

5A and 5B are schematic views showing an example of a photographed image and a method map corresponding thereto. FIG. 5A shows a photographed image 200, and FIG. 5B shows a method map 250 corresponding to the photographed image 200. It is shown in tabular format. For each of the density estimation means 50 and the unit identification means 51, a preliminary experiment using test images of various resolutions revealed the limit (limit resolution) of the resolution of a human image that can be recognized with an accuracy equal to or higher than the lower limit. ing. Further, the density estimation means 50 and the unit identification means 51 are given "A" and "B" in advance as codes (recognition means ID) for identifying them.

In the example of FIG. 5, the area 201 in the photographed image 200 is an area where a person who exists far from the photographing unit 2 can be photographed. When the estimation window is set at an arbitrary position in the region 201, the resolution of the image of a person in the estimation window does not fall below the limit resolution of the density estimation means 50, so that the density estimation result by the density estimation means 50 is obtained. Obtained with an accuracy equal to or higher than the lower limit. Further, when the identification extraction window is set at an arbitrary position in the area 201, the resolution of the image of a person in the identification extraction window is lower than the limit resolution of the single identification means 51, and the identification result by the single identification means 51 is the lower limit. It cannot be obtained with an accuracy higher than the value.

Corresponding to the above, the method map 250 has a reference numeral A assigned to the density estimation means 50 in association with _{the coordinates [..., (X 2} , Y _{2), ...] Of the pixel group in the region 201.} It is remembered.

Further, in the example of FIG. 5, the area 202 in the photographed image 200 is an area where a person existing in the vicinity of the photographing unit 2 can be photographed. When the estimation window is set at an arbitrary position in the region 202, the resolution of the image of a person in the estimation window does not fall below the limit resolution of the density estimation means 50, so that the density estimation result by the density estimation means 50 is obtained. Obtained with an accuracy equal to or higher than the lower limit. Further, when the identification extraction window is set at an arbitrary position in the area 202, the resolution of the image of a person in the identification extraction window does not fall below the limit resolution of the single identification means 51, so that the identification result by the single identification means 51 Is obtained with an accuracy equal to or higher than the lower limit.

Corresponding to the above, the method map 250 includes the reference numeral A assigned to the density estimation means 50 in association with _{the coordinates [..., (X 3} , Y _{3), ...] Of the pixel group in the region 202.} The reference numeral B assigned to the unit identification means 51 is stored.

The distribution estimation means 52 is an object in the target space from the number of objects recognized by the accuracy guarantee recognition means stored in association with the position in the method map storage means 42 among the number of objects recognized by the plurality of recognition means for each position. The distribution information of is generated, and the generated distribution information is output to the distribution information output means 31.

Specifically, the distribution estimation means 52 includes information on the density of the object for each position input from the density estimation means 50, information on the presence or absence of an object for each position input from the single identification means 51, and a method map storage means. The distribution image is generated based on the number recognized by the accuracy guarantee recognition means with reference to the method map stored in 42.

For example, the distribution estimation means 52 projects an object model colored with a color corresponding to the estimated density of the position at a pixel position associated with the reference numerals A and B in the method map and recognized as having an object. Further, a distribution image is generated by setting the color corresponding to the estimated density to the value of the pixel at the position associated only with the reference numeral A representing the density estimation means 50 in the method map. The color can be, for example, red if the estimated density is high, yellow if the estimated density is medium, green if the estimated density is low, or the like.

In generating this distribution image, the distribution estimation means 52 uses the object model stored in the object model storage means 43. The object model storage means 43 stores an object model in which the shape of the object is approximated in advance. The object model is a three-dimensional model composed of three spheroids corresponding to the head, torso, and legs of a standing person. By the way, the center of gravity of the head is the representative position of a person. The three-dimensional model can be simplified so that the entire person is represented by one spheroid, or in more detail, for example, the human head, torso, arms, and legs are represented by separate spheroids. It can also be represented by.

The projection of the object model is basically the same as the projection means 53 of the method map generation process described later, using the object model read from the object model storage means 43 and the camera parameters read from the camera parameter storage means 44. Will be done.

FIG. 6 is a schematic diagram showing an example of a distribution image corresponding to the captured image 200 of FIG. As described above, the reference numeral A and the reference numeral B are associated with the region 202 of FIG. 5, and in FIG. 6, the object whose head center of gravity is placed at the pixel position recognized as the existence of the object corresponding to the region. The projected images 210 to 212 of the model are drawn in colors corresponding to the estimated density of the pixel positions. Here, the hatching of each projected image represents a color corresponding to the estimated density, and the shading of the projected image 210 represents the color corresponding to the high estimated density (red in the above example). Similarly, the horizontal line hatching of the projected image 211 and the diagonal line hatching of the projected image 212 represent colors corresponding to medium density and low density (yellow and green in the above example, respectively).

As described above, only the reference numeral A is associated with the region 201 of FIG. 5, and each pixel of the region 220 of FIG. 6 corresponding to the region is displayed in a color corresponding to the estimated density. Specifically, the hatching of the partial regions 221 to 223 in the region 220 represents the same color as the projection images 210 to 212 described above, the partial region 221 has a high density, the partial region 222 has a medium density, and the partial region 223. Is the region estimated to be low density.

The distribution estimation means 52 may generate a distribution image by transparently synthesizing the distribution image with the captured image. Alternatively, the distribution estimation means 52 may generate a distribution image by transparently synthesizing the distribution image with a projection image of a three-dimensional model that imitates the topography or a building in the target space.

The distribution information output means 31 transmits the distribution information input from the distribution estimation means 52 to the display unit 8 via the display control unit 7 and displays the distribution information.

(Method map creation process)
Next, among the functions of the object distribution estimation device 1, the method map creation process will be described. When performing the method map creation process (FIG. 3), the storage unit 4 functions as the method map storage means 42, the object model storage means 43, the camera parameter storage means 44, the limit value storage means 45, and the like, and the image processing unit 5 performs. It functions as a projection means 53, an application method selection means 54, and the like. The timing for creating the method map is when the method map is not yet stored, when the visual field is changed, or when the estimated desired area is changed. The estimation desired area is an area in the captured image in which the observer wants the object distribution estimation device 1 to estimate the distribution of the object. When the observer inputs the estimation desired area using the operation unit 6, the information of the area is input to the image processing unit 5 via the display control unit 7 and the communication unit 3.

The projection means 53 refers to the estimated desired area input by using the operation unit 6, reads out the object model from the object model storage means 43 and the camera parameters from the camera parameter storage means 44, and each pixel in the estimated desired area. An object model is projected onto the positions to generate a projected image, and the generated projected image for each pixel position is output to the application method selection means 54.

For example, the projection means 53 uses camera parameters to derive a three-dimensional position corresponding to each pixel position in the estimated desired region on a horizontal plane having a height of 160 cm in a virtual space imitating the target space. The height of 160 cm is a predetermined value as the average height of the head of a person in the target space, corresponding to the fact that the center of gravity of the head is the representative position of the person as described above. Then, the projection means 53 arranges the object model in the virtual space with reference to the derived three-dimensional position, and projects the arranged object model on the photographing surface of the photographing unit 2 by using the camera parameters.

The application method selection means 54 compares the resolution of the image of the object at the pixel position with the limit resolution of each recognition means for each pixel position in the estimation desired region, and recognizes the recognition means in which the resolution of the object image is equal to or higher than the limit resolution. Select as an accuracy guarantee recognition means. That is, the application method selection means 54 refers to the resolution of the projected image for each pixel position input from the projection means 53 and the limit resolution of each of the plurality of recognition means stored in the limit value storage means 45. The accuracy guarantee recognition means at each pixel position in the estimation desired region is selected, and the correspondence relationship between the pixel position and the accuracy guarantee recognition means is stored in the method map storage means 42.

Specifically, the resolution can be expressed by the number of pixels of the image of the object. That is, the larger the number of pixels corresponding to the image of the object, the higher the resolution of the image.

In this case, the number of pixels corresponding to the limit resolution is defined as the limit number of pixels, and the limit value storage means 45 can store the limit number of pixels as the limit resolution. Further, the application method selection means 54 counts the number of pixels of the projected image of the pixel position for each pixel position in the estimation desired region. The count value of the number of pixels becomes the resolution of the image of the object at the pixel position, and the application method selection means 54 compares the count value with the limit number of pixels of each recognition means, and the count value is equal to or more than the limit number of pixels. Is selected as the accuracy assurance recognition means.

FIG. 7 is a schematic diagram showing an example of a method map corresponding to the captured image 200 of FIG. In this example, a method map is created for the estimated desired area 300 surrounded by the alternate long and short dash line. _{The projection means 53 is an object at the position [..., (x 1} , y ₁ ), ..., (x ₂ , y ₂ ), ..., (x ₃ , y ₃ ), ...] of the pixel group in the estimated desired region 300. The model is projected to generate a projected image group [..., projected image 310, ..., projected image 320, ..., projected image 330, ...]. _{The application method selection means 54 derives the resolution [..., r 1} , ..., r ₂ , ..., r ₃ , ...] Of each projection image corresponding to each of these projection images. The limit value storing unit 45, the resolution limit R _A, its resolution limit R _B in association with granted code B to a single identification means 51 is stored in association with the code A given to the density estimating means 50 ing. In this example, the resolution _{r 1} of the projected image 310 at the pixel position _{(x 1, y 1) r} 1 <R A, an r 1 _{<R B,} pixel positions of the projection image 320 of _(x 2, _{y 2)} resolution _{r 2 r 2> R a,} an _{r 2 <R B, r 3} > R a for the pixel position _(x 3, _{y 3)} the resolution _{r 3} of the projected image 330 _of, if it is r 2> _{R B} do. Application method selection means 54, the resolution _{[..., r 1, ...,} r 2, ..., r 3, ...] is compared with each of _{R A} and _{R B, [...,} the accuracy of _(x 1, _{y 1)} guarantee recognition means _{no, ..., (x 2, y} 2) accuracy guaranteed recognition means of the density estimating means _{50, ..., (x 3,} y 3) is guaranteed accuracy recognition means of the density estimating means 50, and a single identification means 51 , ...]. That is, the application method selection means 54 determines that there is no accuracy guarantee recognition means for each pixel position in the area 311 including the _{pixel position (x 1} , y _{1 ) in the estimation desired area 300, while the pixel position (x 2).} For each pixel position in the area 321 including, y ₂ ), the association with the code A is stored in the method map storage means 42, and each pixel position in the area 331 including the _{pixel position (x 3} , y _{3) is stored.} The method map storage means 42 stores the association with the code A and the code B.

[Operation of object distribution estimation device 1]
The operation of the object distribution estimation device 1 will be described with reference to the flow charts of FIGS. 8, 9 and 10.

FIG. 8 is a schematic flow chart of the operation of the object distribution estimation device 1. When the object distribution estimation device 1 starts operation, the photographing unit 2 installed at the event venue photographs the monitoring space at predetermined time intervals, and the photographed images are sequentially addressed to the image analysis center where the image processing unit 5 is installed. Send. Then, the image processing unit 5 basically repeats the operation related to the distribution estimation processing in steps S5 to S10 according to the flow chart of FIG. 8 every time the captured image is received. However, when it is necessary to create a method map, the method map creation process of steps S1 to S4 is performed prior to the distribution estimation process of steps S5 to S10.

That is, when the operation unit 6 inputs the change instruction of the estimation desired area to the image processing unit 5 (when “YES” in step S1), the image processing unit 5 performs the method map creation process (step S4). Further, when there is no method map such as when the operation of the object distribution estimation device 1 is started, or when the field of view of the photographing unit 2 is changed (when “YES” in step S2), the image processing unit 5 is, for example, , The display unit 8 is displayed to request the input of the estimated desired area, waits for the input of the estimated desired area from the operation unit 6 (in the case of "NO" in step S3), and when the estimated desired area is input ( If "YES" in step S3), the method map creation process S4 is performed. Then, in these cases, the image processing unit 5 performs the distribution estimation process of steps S5 to S10 after the method map creation process S4.

On the other hand, if the estimated desired area has not changed, the method map already exists, and the visual field has not changed (when “NO” in steps S1 and S2), the image processing unit 5 performs the method map creation process. The distribution estimation process of steps S5 to S10 is performed by omitting S4.

FIG. 9 is a schematic flow chart of the method map creation process S4, and the method map creation process S4 will be described with reference to FIG.

When the estimated desired area is specified in step S1 or S3, the image processing unit 5 operates as the projection means 53, sequentially sets each pixel in the estimated desired area to the attention position (step S40), and stores the object model. The object model read from the means 43 is projected onto the shooting surface of the shooting unit 2 based on the camera parameters read from the camera parameter storage means 44 (step S41). Next, the image processing unit 5 operates as the application method selection means 54, and the application method selection means 54 calculates the resolution of the projected image of the object generated by the projection means 53 (step S42).

The application method selection means 54 compares the resolution of the projected image of the object at the position with the limit resolution of the recognition means for each of the plurality of recognition means for recognizing the number of objects at the position of interest, and is the accuracy guarantee recognition means. If it is an accuracy guarantee recognition means, it is stored in the method map. Specifically, in the present embodiment, there are two types of recognition means using the density estimator and the single classifier, and each is identified by the above-mentioned recognition means IDs A and B.

The application method selection means 54 reads out the limit resolution of the single classifier stored in association with the reference numeral A from the limit value storage means 45, and compares the resolution of the projected image with the limit resolution. Then, if the resolution of the projected image is equal to or higher than the limit resolution (when “YES” in step S43), the unit classifier is stored in the method map storage means 42 as the accuracy guarantee recognition means of the position of interest (step S44). On the other hand, if the resolution of the projected image is less than the limit resolution (when “NO” in step S43), the single classifier is not used as the accuracy guarantee recognition means of the position of interest, and step S44 is omitted.

Following the determination of the accuracy guarantee recognition means for the single classifier in steps S43 and S44, the determination of the accuracy guarantee recognition means for the density estimator is performed in the same manner in steps S45 and S46. That is, the application method selection means 54 reads out the limit resolution of the density estimator stored in association with the reference numeral B from the limit value storage means 45, compares it with the resolution of the projected image, and limits the resolution of the projected image. If it is equal to or higher than the resolution (when "YES" in step S45), the density estimator is stored in the method map storage means 42 as the accuracy guarantee recognition means of the position of interest (step S46), while the resolution of the projected image is limited. If it is less than the resolution (in the case of "NO" in step S45), the density estimator is not used as the accuracy guarantee recognition means of the position of interest, and step S46 is omitted.

The image processing unit 5 repeats the processes of steps S40 to S46 for all the pixels in the estimation desired region (when “NO” in step S47), and when all the pixels are completed (when “YES” in step S47), The process proceeds to step S5 of FIG.

In step S5, the communication unit 3 operates as the image acquisition means 30, and is in a state of waiting for reception of the captured image from the photographing unit 2. The image acquisition means 30 that has acquired the captured image outputs the captured image to the image processing unit 5.

The image processing unit 5 to which the captured image is input operates as the density estimation means 50, and estimates the density distribution from the captured image (step S6). Specifically, the density estimation means 50 extracts the feature amount for estimation of the attention position with each pixel in the estimation desired region in the captured image as the attention position, and estimates the density from the density estimator storage means 40 of the storage unit 4. The instrument is read out, and the feature quantity for estimation is input to the density estimator to obtain the estimated density at the position of interest.

Further, the image processing unit 5 operates as a single identification means 51 and identifies the presence or absence of an object from the captured image (step S7). Specifically, the single identification means 51 extracts the feature amount for identification of the attention position with each pixel in the estimated desired region in the captured image as the attention position, and also extracts the whole body classifier 100, the upper body classifier 101, and the head. Of the neighborhood classifier 102, the single classifier corresponding to the estimated density of the attention position obtained in step S6 is read out from the single body classifier storage means 41, and the identification feature amount is input to the single body classifier to indicate the presence or absence of an object. Identify.

For each pixel in the estimation desired region, the estimated density obtained in step S6 and the presence / absence of an object obtained in step S7 are input to the distribution estimation means 52.

The distribution estimation means 52 reads out the method map stored in the method map storage means 42 (step S8), and generates a distribution image from the method map and the input estimated density and the presence / absence of an object (step S9).

FIG. 10 is a schematic flow chart of the distribution image generation process S9, and the distribution image generation process S9 will be described with reference to FIG. The distribution estimation means 52 sets each pixel in the desired estimation region to the position of interest in order from the distant pixel (step S90). Then, with reference to the method map read out in step S8, it is checked whether the accuracy of the single identification means is guaranteed for the position of interest (step S91).

When the single identification means is the accuracy guarantee recognition means (when "YES" in step S91), the distribution estimation means 52 determines that the identification result of step S7 for the attention position has an object (in step S92, "YES"). If YES ”), a projected image of the object model at the position of interest is drawn (step S93). Here, in the example of the distribution map of the present embodiment described with reference to FIG. 5, the density estimation means is also the accuracy guarantee recognition means at the pixel position where the single identification means is the accuracy guarantee recognition means. Therefore, in step S93, the projected image to be drawn is drawn in a color corresponding to the estimated density in step S6, as described with reference to FIG. On the other hand, if the identification result in step S7 is no object (“NO” in step S92), step S93 is omitted.

If the accuracy of the unit identification means is not guaranteed (in the case of "NO" in step S91), the distribution estimation means 52 refers to the method map and checks whether the density estimation means is accuracy-guaranteed for the position of interest (step). S94).

When the density estimation means is an accuracy guarantee recognition means (when “YES” in step S94), the distribution estimation means 52 imparts a color corresponding to the estimated density in step S6 to the pixels at the position of interest in the distribution image (when the density estimation means is “YES” in step S94). Step S95). On the other hand, if the accuracy of the density estimation means is not guaranteed (“NO” in step S94), step S95 is omitted.

The distribution estimation means 52 repeats the processes of steps S90 to S95 for all the pixels in the desired estimation region (when “NO” in step S96), and when all the pixels are completed (when “YES” in step S96), The process proceeds to step S10 of FIG.

By setting the attention positions in order from the distant pixels in step S90, the hidden surface erasure in the distributed image is performed by the recoating method.

In step S10, the communication unit 3 operates as the distribution information output means 31, outputs the distribution image generated by the distribution estimation means 52 to the display control unit 7, and the display control unit 7 displays the distribution image on the display unit 8. ..

[Modification example]
(1) In the above embodiment, the number of pixels constituting the image of the object is defined as the resolution of the image of the object, but the size of the circumscribing rectangle of the image of the object (that is, the number of pixels or the area) and the height of the image of the object. It is also equivalent to define the size of a rectangle with a certain aspect ratio based on the resolution or the size of a rectangle with a certain aspect ratio based on the width of the image of an object as the resolution. By defining in this way, it becomes easy to derive the resolution of the test image.

(2) In the above embodiment and its modification, the number of pixels or the area is defined as the resolution of the object image, but when the photographing unit 2 and its photographing magnification are determined, the distance and resolution from the photographing unit 2 to the object. Since the relationship with is uniquely determined, the resolution can be estimated from the distance. That is, the distance from the photographing unit 2 to the object can be used as a value representing the resolution of the object image.

In this case, the limit value storage means 45 stores the distance (limit distance) corresponding to the limit resolution. The projection means 53 derives a three-dimensional position corresponding to each pixel position in the estimated desired region and calculates the distance from the photographing unit 2 to the three-dimensional position, and the application method selection means 54 is the distance calculated by the projection means 53. And the limit distance are compared to select the accuracy guarantee recognition means. That is, the application method selection means 54 selects, among the plurality of recognition means, the one whose distance to the object photographed at the position is equal to or less than the limit distance as the accuracy guarantee recognition means with respect to the position in the captured image. .. Then, the method map storage means 42 stores the position in the captured image in association with the accuracy guarantee recognition means.

For example, the projection means 53 back-projects the position in the captured image onto a horizontal plane having a height corresponding to the object in the target space to obtain the virtual position of the object, and the application method selection means 54 corresponds to the position in the captured image. The distance between the virtual position of the object corresponding to the position and the photographing unit 2 is compared with the limit distance of each of the plurality of recognition means, and the accuracy guarantee recognition means associated with the position is selected.

(3) In the above embodiment and its modification, the distribution estimation means 52 generates a distribution image as distribution information, but the distribution information is not limited to image representation. For example, the distribution estimation means 52 determines the position of a pixel in which the single identification means 51 recognizes the existence of an object in the captured image among the pixels associated with the single identification means 51 as the accuracy guarantee recognition means in the distribution map. The information of the three-dimensional position where each object exists is calculated by back-projecting to the target space according to the camera parameters of the photographing unit 2. Further, in the distribution estimation means 52, for the pixels to which the density estimation means 50 is associated as the accuracy guarantee recognition means in the distribution map, the pixels whose density is estimated in the captured image are grouped into regions for the same estimated density, and each region. Is back-projected onto the target space according to the camera parameters of the photographing unit 2 to calculate the information of the three-dimensional region of each density. In this way, the distribution estimation means 52 may generate three-dimensional distribution information. Further, in this case, with respect to the position to which both the density estimation means 50 and the unit identification means 51 are associated, the distribution information may be generated only from the more detailed information of the unit identification means 51.

(4) In the above-described embodiment and its modification, an example in which the object to be detected is a human is shown, but the object to be detected may be a vehicle, an animal such as a cow or a sheep, or the like. can.

(5) In the above embodiment and each modification thereof, the density estimator learned by the multi-class SVM method is illustrated, but instead of the multi-class SVM method, a decision tree type random forest method and multi-class AdaBoost are used. Various density estimators such as density estimators learned by the (AdaBoost) method or the multiclass logistic regression method can be used.

Alternatively, it can be a density estimator using an identification type CNN (Convolutional Neural Network).

(6) In the above-described embodiment and each modification thereof, the classes of densities other than the background estimated by the density estimator are set to 3 classes, but the classes may be further divided.

(7) In the above-described embodiment and each modification thereof, a density estimator classified into multiple classes is illustrated, but instead of this, a regression-type density estimator that returns a density value (estimated density) from a feature amount. It can also be. That is, with a density estimator that has learned the parameters of the regression function for obtaining the estimated density from the features by the ridge regression method, the support vector regression method, the random forest method of the regression tree type, or the Gaussian Process Regression. can do.

Alternatively, it can be a density estimator using a regression type CNN.

(8) In the above-described embodiment and each modification thereof, the GLCM feature is illustrated as the feature amount learned by the density estimator and the feature amount for estimation, but these are replaced with the GLCM feature and are local binary patterns (Local Binary). Pattern: LBP) features, Haar-like features, HOG features, brightness patterns, etc. can be used, or GLCM features and a plurality of these can be combined. You can also do it.

(9) In the above-described embodiment and each modification thereof, an example is shown in which the density estimation means 50 and the unit identification means 51 scan at 1-pixel intervals to perform processing, but these scans are performed at intervals of 2 pixels or more. It is also possible to do it in the open.

(10) In the above-described embodiment and each modification thereof, a single classifier learned by the linear SVM method is illustrated, but various conventionally known learning methods such as the AdaBoost method are used instead of the linear SVM method. It can also be a single classifier learned using. In addition, a pattern matching device can be used instead of the classifier, and the discrimination score in that case is the inner product of the average pattern of the feature amount extracted from the human learning image and the feature amount of the input image, and the discrimination score is calculated. The function can be a function in which the score is used as an output value and the feature amount of the input image is used as an input value. Further, an identification type CNN may be used as a single classifier.

(11) In the above-described embodiment and each modification thereof, the HOG feature amount is exemplified as the feature amount learned by the single classifier, but these are the local binary pattern feature amount and the Haar-like feature amount instead of the HOG feature amount. , Brightness pattern and the like, or a combination of the HOG feature amount and a plurality of these can be used.

1 Object distribution estimation device, 2 Imaging unit, 3 Communication unit, 4 Storage unit, 5 Image processing unit, 6 Operation unit, 7 Display control unit, 8 Display unit, 30 Image acquisition means, 31 Distribution information output means, 40 Density estimation Instrument storage means, 41 single-unit classifier storage means, 42 method map storage means, 43 object model storage means, 44 camera parameter storage means, 45 limit value storage means, 50 density estimation means, 51 single-unit identification means, 52 distribution estimation means, 53 Projection means, 54 Applicable method selection means, 100 Whole body classifier, 101 Upper body classifier, 102 Head proximity classifier, 200 Captured image, 250 Method map, 300 Estimated desired area.

Claims (6)

An image acquisition means for acquiring a photographed image of a space in which a predetermined object photographed by the photographing unit may exist, and
A recognition means for recognizing the number of objects photographed at an arbitrary position in the photographed image by using a recognizer in which the characteristics of each of a plurality of types of learning images in which the number of captured objects is different are learned in advance. A plurality of recognition means provided for each range of the number of objects by using the plurality of recognizers having different ranges of the number of recognizable objects.
A storage means for storing the position in the captured image in association with the accuracy guarantee recognition means capable of recognizing the number of objects at the position with an accuracy equal to or higher than a predetermined lower limit value among the plurality of recognition means.
A distribution estimation means that generates distribution information of the object in the space from the number of objects acquired in the captured image by the accuracy guarantee recognition means stored in the storage means.
An object distribution estimation device characterized by being equipped with. The lower limit of the resolution of the image of the object that can be recognized by the recognition means with an accuracy equal to or higher than the lower limit is defined as the limit resolution of the recognition means.
The storage means associates the plurality of recognition means having the resolution of the image of the object at the position equal to or higher than the limit resolution as the accuracy guarantee recognition means with respect to the position in the captured image. I remember
The object distribution estimation device according to claim 1. The resolution is represented by the number of pixels of the image of the object, and the number of pixels corresponding to the limit resolution is defined as the limit number of pixels.
A projection means that generates a projected image by projecting a model that imitates the object existing in the space onto the imaging surface of the imaging unit.
The accuracy guarantee recognition means associated with the position by comparing the number of pixels of the projected image generated at the position with the limit number of pixels of each of the plurality of recognition means with respect to the position in the captured image. Selection means to select and
The object distribution estimation device according to claim 2, further comprising. The resolution is estimated from the distance from the photographing unit to the object in the space, and the distance corresponding to the limit resolution is set as the limit distance.
The storage means uses a plurality of recognition means whose distance with respect to the object photographed at the position is equal to or less than the limit distance with respect to the position in the captured image as the accuracy guarantee recognition means. What you associate and remember,
2. The object distribution estimation device according to claim 2. A projection means for obtaining a virtual position of an object by back-projecting a position in the captured image onto a horizontal plane having a height corresponding to the object in the space.
The accuracy associated with the position by comparing the distance between the virtual position of the object corresponding to the position and the photographing unit and the limit distance of each of the plurality of recognition means with respect to the position in the captured image. Selection means for selecting guarantee recognition means and
The object distribution estimation device according to claim 4, further comprising. An image acquisition means for acquiring a photographed image of a space in which a predetermined object photographed by the photographing unit may exist, and
The density of the object photographed at an arbitrary position in the photographed image using a density estimator that has learned in advance the characteristics of each density image obtained by photographing the space in which the object exists at each predetermined density. Density estimation means to recognize
A single identification means for recognizing the presence or absence of the object photographed at an arbitrary position in the captured image by using a single classifier in which the characteristics of the single image in which the single object is photographed are learned in advance.
The position in the captured image is associated with the accuracy guarantee recognition means that can recognize the density or presence / absence of the object at the position among the density estimation means and the single unit identification means with an accuracy equal to or higher than a predetermined lower limit value. Memories to memorize
A distribution estimation means that generates distribution information of the object in the space from the density or presence / absence of the object acquired in the captured image by the accuracy guarantee recognition means stored in the storage means.
An object distribution estimation device characterized by being equipped with.

Images