JP2014006655A - Passenger number estimation method and computer program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce a calculation cost required for passenger number estimation.SOLUTION: An estimation method comprises: an acquisition step of acquiring the number of moving bodies from an image in time series; a staying time estimation step of estimating a staying time during which each of the moving bodies has stayed in an imaging area; and a passage estimation step of estimating the number of moving bodies that have passed by using the number of the moving bodies acquired in the acquisition step and the staying time of the moving bodies estimated in the staying time estimation step.

Description

  The present invention relates to a technique for estimating the number of people passing by.

A technique for measuring the number of people passing through a passage using a camera image is required from the viewpoint of marketing and safety management support, and has already been put into practical use in part. Various image processing techniques for realizing these techniques for measuring the number of passing people have been proposed, and can be roughly divided into two approaches.
The first approach is based on a person tracking technique (tracking technique) in which an individual person existing on an image is detected and its position is captured and tracked. However, in the person tracking technique, there is a problem that it becomes difficult to detect and track individual persons due to occurrence of occlusion (an object is hidden behind another object in the image). Such a problem occurs, for example, when a person tries to be applied to a place where a lot of people are crowded, such as a concourse of a large station, and people overlap each other on an image.

  The second approach is a technique in which an individual person is not detected but the whole group of persons is regarded as a group like a fluid (see Non-Patent Document 1). In this method, the number of passing people is estimated by calculating the number of people in the region at the moment and the movement amount (movement speed) of the entire group within a certain time. Such an approach that does not require the detection of individual persons can be expected to operate stably even when it is congested.

Isamu Igarashi, 3 others, “Measurement of the number of people passing from video using collective movement based on voting regarding local movement”, IS3-03, 2011 Image Sensing Symposium, 2011

In the method of Non-Patent Document 1 described above, person feature points are associated between two temporally adjacent frame images, and a representative velocity of the entire group is estimated based on the result of the association. Was. In addition, the number of passing people is estimated by estimating the estimated representative speed and the number of people in the area at the moment.
However, with this method, there is a problem that when the number of feature points increases as the number of people increases, the cost of associating feature points becomes very high.

  In view of the above circumstances, an object of the present invention is to provide a technique for reducing the calculation cost for estimating the number of passing people.

  According to one aspect of the present invention, an acquisition step of acquiring the number of moving bodies from an image in time series, a stay time estimation step of estimating a stay time of the moving body staying in the imaging region, and the acquisition step An estimation method comprising: a passage estimation step of estimating the number of mobile bodies that have passed using the number of mobile bodies and the residence time of the mobile bodies estimated in the residence time estimation step.

  One aspect of the present invention is the above-described estimation method, wherein the acquisition step uses a conversion coefficient that normalizes the size of each pixel of the image to determine a difference in size of the moving object in the image. By normalizing, the number of the moving bodies is acquired in time series.

  One aspect of the present invention is the above estimation method, wherein the obtaining step creates a projected image in which foreground pixels are projected in one dimension using a conversion coefficient that normalizes the size of each pixel of the image. The number of the moving bodies within a predetermined time is calculated using the projected image, and the staying time estimating step uses the projected image generated at the acquisition step to use the staying time of the moving body within the predetermined time. To get.

  According to one aspect of the present invention, an acquisition step of acquiring the number of moving bodies from an image in time series, a stay time estimation step of estimating a stay time of the moving body staying in the imaging region, and the acquisition step A computer program for causing a computer to execute a passage estimation step of estimating the number of mobile bodies that have passed using the number of mobile bodies and the stay time of the mobile bodies estimated in the stay time estimation step It is.

  According to the present invention, it is possible to reduce the calculation cost for estimating the number of passing people.

It is a figure which shows the system configuration | structure of a passage number estimation system. It is a schematic block diagram showing the functional composition of passing number estimating device 100 of the present invention. It is a figure showing the method of calculating the scale conversion coefficient w (x, y). It is a figure which shows the mode of the process of a one-dimensional projection. It is a figure showing an example of the process of a one-dimensional projection. It is a figure showing the specific example of the image produced | generated according to the direction of an optical flow. It is a figure which shows the other specific example of a projection image. It is a flowchart which shows the flow of the process of the initial setting of this invention. It is a flowchart which shows the flow of the process of estimation of the number of passing persons of this invention.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a diagram showing a system configuration of a passing person estimation system. The passing person estimation system of the present invention includes a fixed camera 10 and a passing person estimation device 100.
The fixed camera 10 is installed obliquely downward like a security camera, and images the imaging region 20. In addition, the fixed camera 10 captures an image so that the motion of the person on the image has a lateral motion component by capturing the image from the side or diagonal direction with respect to the moving direction of the person such as the passage. That is, the fixed camera 10 is installed so that a person moves less toward the fixed camera 10 in a direction close to the front.

The passing person estimation device 100 is configured using an information processing device. The passing number estimating device 100 estimates the number of people who have passed through the shooting area 20 by performing image processing based on the frame image shot by the fixed camera 10.
The shooting area 20 is an area shot by the fixed camera 10. Reference numerals 30-1 to 30-3 represent a person moving in the imaging region 20. Arrows 40-1 to 40-3 indicate the traveling direction in which the persons 30-1 to 30-3 are moving.

  FIG. 2 is a schematic block diagram showing a functional configuration of the passing person estimation device 100 of the present invention. The passing person estimation device 100 includes a CPU (Central Processing Unit), a memory, an auxiliary storage device, and the like connected by a bus, and executes a passing person estimation program. By executing the passing number estimation program, the passing number estimation apparatus 100 includes a frame image input unit 101, a scale conversion coefficient storage unit 102, a foreground detection unit 103, an optical flow detection unit 104, a foreground division unit 105, a one-dimensional projection unit 106, It functions as an apparatus including a projected image generation unit 107, a flow parameter estimation unit 108, and a passing person estimation unit 109. Note that all or part of the functions of the passing person estimation device 100 may be realized using hardware such as an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a field programmable gate array (FPGA). good. The passing number estimation program may be recorded on a computer-readable recording medium. The computer-readable recording medium is, for example, a portable medium such as a flexible disk, a magneto-optical disk, a ROM, a CD-ROM, or a storage device such as a hard disk built in the computer system. Further, the passing number estimation program may be transmitted / received via a telecommunication line.

The frame image input unit 101 inputs frame images taken by the fixed camera 10 in chronological order.
The scale conversion coefficient storage unit 102 is configured using a storage device such as a magnetic hard disk device or a semiconductor storage device. The scale conversion coefficient storage unit 102 stores a scale conversion coefficient w (x, y) corresponding to each pixel of the frame image captured by the fixed camera 10.

The foreground detection unit 103 detects foreground pixels in the frame image input by the frame image input unit 101. A foreground pixel is a pixel relating to an object (here, mainly a person) that temporarily appears on the spot. On the other hand, a pixel relating to an object that is regularly present at the place is called a background pixel.
Detection of foreground pixels is a technique that has existed for a long time in the field of image processing, and various methods have already been proposed. For example, the foreground detection unit 103 acquires the average value of each pixel for a certain past time (for example, about 5 minutes) as the background pixel value. The foreground detection unit 103 calculates the absolute value of the difference between each pixel value of the current frame image and the background pixel value, and detects a pixel whose value is greater than or equal to the threshold as a foreground pixel. The above process is merely an example of the foreground pixel detection process, and other detection processes may be used.

The optical flow detection unit 104 detects an optical flow in temporally adjacent frame images input by the frame image input unit 101. Hereinafter, an example of the optical flow detection process will be described. The optical flow detection unit 104 detects a feature point of a person. The optical flow detection unit 104 associates the feature points of the person and detects the optical flow. The optical flow detection unit 104 detects a motion in each pixel of the frame image by detecting an optical flow for each pixel.
Through the above processing, the optical flow detection unit 104 acquires the velocity fields Vx (x, y) and Vy (x, y) on the frame image. Vx represents a velocity component in the x-axis direction, and Vy represents a velocity component in the y-axis direction. These velocity fields Vx (x, y) and Vy (x, y) are calculated for each pixel (x, y).

The foreground division unit 105 performs division processing using the foreground pixels detected by the foreground detection unit 103 and the optical flow detected by the optical flow detection unit 104. In the dividing process, the foreground dividing unit 105 classifies the foreground pixels according to the direction of the optical flow (for example, the left direction and the right direction), and generates a left direction image and a right direction image. The left direction image is an image in which only a foreground pixel in the left direction (hereinafter referred to as “left foreground pixel”) is given a pixel value different from the background pixel value. The right direction image is an image in which only a foreground pixel in the right direction (hereinafter referred to as “right foreground pixel”) is given a pixel value different from the background pixel value.
The one-dimensional projection unit 106 uses each image (left direction image and right direction image) divided by the foreground division unit 105 and the scale conversion coefficient w (x, y) on a predetermined axis (for example, the x axis). A corresponding projection value p (x) is calculated.

The projected image generation unit 107 acquires a projected value p (x) corresponding to a frame image for a predetermined time. The projected image generation unit 107 generates a projected image P (x, t) by arranging the acquired projection values p (x) in time series.
The flow parameter estimation unit 108 estimates the flow parameter based on the projection image P (x, t) generated by the projection image generation unit 107. The flow parameter is a parameter related to the movement of the person within the imaging region. The flow parameter is, for example, the moving speed of the person, the staying time of the person, and the like.
The passing number estimating unit 109 uses the flow parameter estimated by the flowing parameter estimating unit 108 to estimate the number of passing through the shooting area in the input frame image for a predetermined time.

Next, a method for calculating the scale conversion coefficient w (x, y) will be described with reference to FIG.
FIG. 3 is a diagram illustrating a method for calculating the scale conversion coefficient w (x, y). FIGS. 3A to 3C show specific examples of images taken in a state where a person is located near, away from, and further away from the fixed camera 10, respectively. As described above, on the image captured by the fixed camera 10, a person located near the fixed camera 10 occupies a large area on the image, and a person located away from the fixed camera 10 is small on the image. Occupies an area. The set of foreground pixels for one person in FIG. 3A is represented as set f, the set of foreground pixels for one person in FIG. 3B is represented as set f1, and the set of foreground pixels for one person in FIG. Is represented as a set f2. In this case, it is possible to define s (x, y) (hat is written on s. The same applies hereinafter) such that Equation 1 holds (Reference 1: Hiroyuki Arai, 3 others, “Counting people from video” "Study on geometric invariants for"", Eijiji Technical Report, vol.34, no.45, ME2010-163, pp.41-45).

  The sigma in Equation 1 represents the sum of all the pixels (x, y) in the frame image. When the scale conversion coefficient w (x, y) is expressed as in Expression 2 using the above s (x, y), Expression 3 is established.

  Note that a specific method for calculating the scale conversion coefficient w (x, y) is not necessarily based on Reference Document 1. For example, the scale conversion coefficient w (x, y) may be calculated by the following method. First, quantitatively estimate how much surface area each pixel of the frame image photographed by the fixed camera 10 corresponds to in the vicinity of the height of the person in the real space (the extent of the pixel in the real space). To do. It is also possible to approximately obtain the scale conversion coefficient w (x, y) by performing the above-described estimation.

  The scale conversion coefficient w (x, y) is calculated in advance for each pixel using camera parameters. The scale conversion coefficient storage unit 102 stores the calculated scale conversion coefficient w (x, y) in advance. The camera parameters are internal parameters and external parameters of the fixed camera 10. For example, the internal parameter is a parameter related to the configuration of the fixed camera 10. Internal parameters represent focal length, aspect ratio, distortion parameters, and the like. External parameters are parameters relating to the position (including height) and orientation of the fixed camera 10 in real space. Each camera parameter is calculated by calibrating the fixed camera 10.

  In camera calibration, internal parameters and external parameters are estimated. In the estimation of the internal parameters, the optical system (focal length, aspect ratio, distortion parameter, etc.) of the fixed camera 10 is obtained. In the estimation of the external parameters, the three-dimensional position (X, Y, Z) of the fixed camera 10 in the real space and the shooting direction (θ, φ, ω) of the fixed camera 10 are obtained. In the three-dimensional position (X, Y, Z), for example, X and Y represent a two-dimensional position on the imaging region 20, and Z represents a height from the floor surface. In the shooting direction (θ, φ, ω) of the fixed camera 10, for example, θ represents rotation around the x axis, φ represents rotation around the y axis, and ω represents rotation around the z axis. Based on both estimation results, it is determined in advance which position in the real space each pixel of the frame image is photographed.

  Various techniques exist for camera calibration for calculating the internal parameters and the external parameters of the fixed camera 10. Any technique may be applied in the present embodiment.

FIG. 4 is a diagram illustrating a state of the one-dimensional projection process. Hereinafter, the one-dimensional projection process will be described with reference to FIG.
FIG. 4A shows a scale conversion coefficient image. The scale conversion coefficient image is an image in which the scale conversion coefficient w (x, y) is recorded as a pixel value in each pixel. The upper diagram of FIG. 4B is a diagram showing the foreground image (assuming that the pixel corresponding to the human body is the foreground pixel). A foreground image is an image in which different pixel values are assigned to foreground pixels and background pixels. For example, when “1” is assigned to the foreground pixel and “0” is assigned to the background pixel, each pixel value F (x, y) of the foreground image is expressed as in Expression 4 below.

  The lower diagram of FIG. 4B is a diagram illustrating a specific example of the result of the one-dimensional projection. The vertical axis in the lower diagram of FIG. 4B represents the projection value p (x), and the horizontal axis represents the x-axis.

  The one-dimensional projection unit 106 generates a foreground coefficient image by multiplying each pixel value of the foreground image in the upper diagram of FIG. 4B and the scale conversion coefficient image in FIG. 4A. The pixel value of each pixel of the foreground coefficient image is a result of multiplication of w (x, y) and F (x, y). That is, only the foreground pixel has the scale conversion coefficient w (x, y), and the background pixel has “0”. The one-dimensional projection unit 106 performs one-dimensional projection on the foreground coefficient image. In the present embodiment, the one-dimensional projection unit 106 performs a one-dimensional projection of the foreground coefficient image on the x axis. The value of each X coordinate obtained by the one-dimensional projection is the projection value p (x). Therefore, the projection value p (x) is expressed as in Equation 5. Note that the sigma in Expression 5 is the sum related to the coordinate y in the height direction of the image.

  In the above-described processing, the axis projected by the one-dimensional projection unit 106 is not necessarily on the x-axis, and projection onto an arbitrary straight line on the image is possible. By performing the processing after rotating the image by a predetermined angle in advance, a one-dimensional projection on an arbitrary straight line can be performed by the same processing as the projection onto the x-axis. Which direction is selected as the projection axis may be determined according to the moving direction of the person in the image. When the fixed camera 10 takes a picture of the passage from the horizontal direction, the movement of the person on the image moves in the left-right direction. Therefore, projection on the x-axis (the horizontal direction of the image) is desirable.

  FIG. 5 is a diagram illustrating an example of a one-dimensional projection process. FIG. 5A shows a frame image taken by the fixed camera 10. The horizontal axis represents the x axis, and the vertical axis represents the y axis. FIG. 5B shows the foreground image of the frame image shown in FIG. FIG. 5C shows a specific example of the result of the one-dimensional projection.

FIG. 6 is a diagram illustrating a specific example of an image generated according to the direction of the optical flow. FIG. 6A is a diagram illustrating a specific example of the direction image generated by assigning different pixel values to the foreground pixels according to the direction of the optical flow. In other words, FIG. 6A shows an image generated by giving different pixel values to the left foreground pixel, the right foreground pixel, and the background pixel, respectively.
In FIG. 6A, a first value is given as a pixel value to a foreground pixel (right foreground pixel) in which a movement in the right direction (ie, positive direction) is observed in the frame image, and left direction (ie, negative). The second value is given as the pixel value to the foreground pixel (left foreground pixel) in which movement in the direction) is observed, and the third value is given as the pixel value to the background pixel. When the pixel value of the right direction image is defined as R (x, y) and the pixel value of the left direction image is defined as L (x, y), each pixel value is calculated by Expression 6.

  The one-dimensional projection unit 106 generates a foreground coefficient image by using a right direction image and a left direction image as the foreground image. Specifically, the one-dimensional projection unit 106 generates a right foreground coefficient image by multiplying each pixel value of the right direction image and the scale conversion coefficient image. The one-dimensional projection unit 106 generates a left foreground coefficient image by multiplying each pixel value of the left direction image and the scale conversion coefficient image. The pixel value of each pixel of the right foreground coefficient image is a result of multiplication of w (x, y) and R (x, y). The pixel value of each pixel of the left foreground coefficient image is the result of multiplication of w (x, y) and L (x, y).

  FIG. 6B is a diagram illustrating a specific example of a result of one-dimensional projection of the left foreground coefficient image. FIG. 6C is a diagram illustrating a specific example of the result of the one-dimensional projection of the right foreground coefficient image. FIG. 6D is a diagram illustrating a specific example of the projected image P (x, t). The projection image P (x, t) is generated by arranging the projection values p (x) of the left foreground coefficient image and the right foreground coefficient image in time series. The horizontal axis x of the projected image P (x, t) is the x-axis of the frame image, and the vertical axis t represents the time taken by the fixed camera 10. The projection image for the projection value p (x) of the right foreground coefficient image is represented as Pr (x, t), and the projection image for the projection value p (x) of the left foreground coefficient image is represented as Pl (x, t). The flow parameter estimation unit 108 estimates the representative moving speed of the person by estimating the slope of the line formed by Pr (x, t) or Pl (x, t).

  FIG. 7 is a diagram showing another specific example of the projected image. Hereinafter, the process of estimating the number of passing people will be described with reference to FIG. The projected image Pl (X, T) is used. The horizontal axis X of the projected image Pl (X, T) is the x-axis of the frame image, and the vertical axis T represents the time taken by the fixed camera 10. The flow parameter estimation unit 108 obtains the slope of the line of the projection image Pl (X, T). As a method for obtaining the inclination of this line, there is a method by detecting the gradient of the projected image Pl (X, T). The flow parameter estimation unit 108 calculates the gradient direction (the direction of the arrow in FIG. 7) for each pixel of the projected image Pl (X, T). The calculation method of the gradient direction may be calculated by applying a first-order differential filter in the vertical and horizontal directions and taking an arc tangent of each direction component.

The flow parameter estimation unit 108 votes for each calculated gradient direction from each pixel of the projected image Pl (X, T). The vote is added one vote each time the same angle (for example, divided into units of 1 degree) is calculated in the gradient direction calculated from each pixel of the projected image Pl (X, T). It is processing. The flow parameter estimation unit 108 estimates the mode value in the gradient direction based on the voted result.
The flow parameter estimation unit 108 sets the estimated mode value of the gradient direction as the gradient direction of the projection image Pl (X, T). The inclination of the line orthogonal to the gradient direction (the line extending from the lower right to the upper left in FIG. 7) is estimated as a representative velocity (hereinafter, referred to as velocity V) of the projected image Pl (X, T). The flow parameter estimation unit 108 estimates the person's stay time τ from the velocity V and the x-axis of the projected image Pl (X, T). The staying time is the time during which a person stays in the shooting area 20. Therefore, the person's stay time τ is expressed as shown in Equation 7.

  The stay time τ obtained by Expression 7 is the time during which a single person who passes through the imaging region 20 at the speed V stays in the frame image for a predetermined time.

  Next, the flow parameter estimation unit 108 obtains the sum Nsum of the projection values p (x) of the entire projection image Pl (X, T) (0 to T) for a predetermined time. The total sum Nsum of the projection values p (x) is expressed as in Expression 8.

  Nsum is the sum of the scale conversion coefficients w (x, y) corresponding to the left foreground pixels based on the leftward image for a predetermined time. That is, Nsum is the result of adding the number of persons passing through the shooting area 20 existing in each left image from all left images taken between the shooting times 0 to T of the fixed camera 10. . From the above results, the passing number estimating unit 109 estimates the number of passing persons (passing number) N in the left direction by using the staying time τ of the person and the total Nsum of the left foreground images for a predetermined time. An equation for estimating the number N of passing people is expressed as Equation 9.

  Based on Equation 9 described above, the passing number estimating unit 109 estimates the passing number N. Further, the passing number estimating unit 109 estimates the passing number for the person moving in the right direction by the same method.

Next, the processing flow of the present invention will be described with reference to FIGS.
FIG. 8 is a flowchart showing the flow of the initial setting process according to the present invention. This initial setting process may be performed once before the number of passing people is measured. The initial setting may be executed, for example, by an administrator of the passing person estimation system. First, the administrator calibrates the fixed camera 10 and calculates camera parameters (step S101). Then, the administrator uses the camera parameter to calculate a scale conversion coefficient w (x, y) that normalizes the difference in size with respect to the position on the image (step S102). The administrator records the calculated scale conversion coefficient w (x, y) in the scale conversion coefficient storage unit 102 of the passing number estimating device 100.

  FIG. 9 is a flowchart showing the flow of the process of estimating the number of passing persons according to the present invention. First, the frame image input unit 101 inputs frame images in time series (step S201). The foreground detection unit 103 detects the foreground pixels of the frame image input by the frame image input unit 101 (step S202). The optical flow detection unit 104 detects feature points of a person from temporally adjacent frame images input by the frame image input unit 101. The optical flow detection unit 104 detects a human feature point from the frame image, and then associates the feature point of the frame image with each other to detect an optical flow. (Step S203).

  The foreground division unit 105 performs division processing using the foreground pixels detected by the foreground detection unit 103 and the optical flow detected by the optical flow detection unit 104. The foreground division unit 105 classifies the foreground pixels according to the direction of the optical flow, and generates a left direction image and a right direction image (step S204). The one-dimensional projection unit 106 generates a foreground coefficient image by multiplying each image divided by the foreground division unit 105 and each pixel value of the scale conversion coefficient w (x, y). Thereafter, the one-dimensional projection unit 106 performs a one-dimensional projection on the foreground coefficient image and obtains a projection value p (x). The projected image generation unit 107 generates a projected image P (x, t) by arranging the projection values p (x) acquired by the one-dimensional projection unit 106 in time series (step S205).

Next, the flow parameter estimation unit 108 determines whether or not a predetermined time has elapsed from the estimation of the flow parameter (step S206). When a predetermined time has elapsed since the estimation of the flow parameter (YES in step S206), the flow parameter estimation unit 108 performs a process of estimating the gradient of the line formed in the projection image P (x, t) (step S207). . The flow parameter estimation unit 108 estimates a representative speed of the person from the estimated gradient of the projected image P (x, t). The flow parameter estimation unit 108 estimates the number of people passing through the imaging region 20 in all frame images for a predetermined time (step S208). The passing number estimating unit 109 estimates the passing number based on the stay time τ of the person estimated by the flow parameter estimating unit 108 and the passing number of all frame images for a predetermined time estimated by the flow parameter estimating unit 108. (Step S209).
On the other hand, when the predetermined time has not elapsed since the estimation of the flow parameter, the passing person estimation system returns to the process of step S201 and repeatedly executes the process of estimating the passing person (NO in step S206). The passing person estimation system repeatedly executes the above processing until an end instruction is input by the user or the power is turned off by the user.

  According to the passing person estimation system configured as described above, a projection image is generated by detecting the moving direction of a person without associating each feature point related to the height of the person. By calculating the gradient of the projection value from the generated projection image, the moving speed V of the representative person can be detected. Then, by detecting the moving speed V of a representative person, the number of passing people can be estimated. Therefore, it is possible to reduce the calculation cost for associating feature points when estimating the number of passing people.

<Modification>
The fixed camera 10 and the passing person estimation device 100 may be configured integrally.
The frame image captured by the fixed camera 10 may be once recorded on a recording medium such as a hard disk, and the passing number estimating device 100 may perform the processing while reading the frame image.
When the passing person estimation device 100 estimates the passing person for a long time (for example, a time width T such as 1 hour), the passing person number between the time widths T is repeatedly measured by the method described above (1). The measurement may be performed 60 times per minute and the results are added), or the time width T may be 1 hour, and the measurement may be performed at once.
In addition, when the passing number estimation device 100 measures the time width T as 1 hour at a stretch, in a place where the flow state (speed and number of people) changes greatly during the time width T, the passing time estimation apparatus 100 is repeated at a short time interval t. It is desirable to measure.
The foreground detection unit 103 may acquire an average value in the movement direction of each pixel as a background pixel value.

  The embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to this embodiment, and includes designs and the like that do not depart from the gist of the present invention. An example is described below. In a place where the flow direction of a person is fixed in one direction, in the above description, it is possible to estimate the number of passing people without detecting the optical flow and classifying (dividing) the foreground image by the optical flow. It is. In this case, the projected image P (x, t) in FIG. 6D is a projected image P (x, t) of the foreground image itself, and is an image that is not classified (divided) according to the left and right directions. It is obvious that the gradient (stay time τ) can be calculated for this image by voting as in the above example.

DESCRIPTION OF SYMBOLS 10 ... Fixed camera, 20 ... Shooting area | region, 30-1-3 ... person, 40-1-3 ... arrow (traveling direction), 100 ... Passing number estimation apparatus, 101 ... Frame image input part, 102 ... Scale conversion coefficient memory | storage , 103 ... Foreground detection unit, 104 ... Optical flow detection unit, 105 ... Foreground division unit, 106 ... One-dimensional projection unit, 107 ... Projection image generation unit, 108 ... Flow parameter estimation unit, 109 ... Passing number estimation unit

Claims (4)

An acquisition step of acquiring the number of moving objects in time series from the image;
A staying time estimating step for estimating a staying time in which the moving object has stayed in the imaging region;
A passage estimation step for estimating the number of the mobile bodies that have passed using the number of the mobile bodies acquired in the acquisition step and the stay time of the mobile bodies estimated in the stay time estimation step;
An estimation method comprising:   The acquisition step uses a conversion coefficient that normalizes the size of each pixel of the image to normalize the difference in size of the moving objects in the image, so that the number of moving objects is time-sequentially. The estimation method according to claim 1 to be acquired. The obtaining step creates a projected image in which foreground pixels are projected in a one-dimensional manner using a conversion coefficient that normalizes the size of each pixel of the image, and uses the projected image of the moving object within a predetermined time. Calculate the number,
The estimation method according to claim 1, wherein the staying time estimating step acquires the staying time of the moving body within a predetermined time using the projection image generated in the acquiring step. An acquisition step of acquiring the number of moving objects in time series from the image;
A staying time estimating step for estimating a staying time in which the moving object has stayed in the imaging region;
A passage estimation step for estimating the number of the mobile bodies that have passed using the number of the mobile bodies acquired in the acquisition step and the stay time of the mobile bodies estimated in the stay time estimation step;
A computer program for causing a computer to execute.