JP6187261B2 - Person feature amount extraction system, server, and person feature amount extraction method - Google Patents

Description

  The present invention relates to a system, an imaging terminal, a server, and a method for extracting a feature amount of a person from an image.

  For purposes such as monitoring, tracking, and searching for a person, it is desired to specify the same person (person identification) between videos of a single camera and multiple cameras. In order to track or search for a person, a feature amount relating to the appearance of the person's face or clothing is extracted from the image. In order to extract the feature amount of the person, it is necessary to extract the person shown in the image and calculate the feature amount for each extracted person.

  For example, since surveillance cameras generally cover a wide range, they are often photographed from a high position using a wide-angle lens or the like. For this reason, the size of the person in the image is small, and the person is often photographed with an angle of view looking down from above. In order to extract a person's head and face from a camera image shot in such an environment, high calculation processing is required. For this reason, since it is difficult to perform processing in a small image processing apparatus installed in the camera, the image captured by the camera is usually transferred to a server capable of high-calculation processing, and person extraction is performed on the server. And feature amount calculation.

  However, when transferring the camera video to a server that performs image processing, if the image itself is sent, a privacy problem arises. For this reason, it is necessary to take measures such as transferring data (features) to the server side after performing image processing on the camera side to convert the original image into information that cannot be restored.

  In the existing technology, Patent Document 1 proposes a method of analyzing a person's behavior without infringing on the person's privacy.

  In the method of Patent Document 1, a numerical value representing a feature of a person shown in a captured video is extracted as feature data and encrypted. Then, the imaging device transmits the encrypted feature amount data, the predetermined related information, and the imaging device identifier uniquely given to the imaging device to the verification device. The collation device stores the encrypted feature amount data, the related information, and the photographing device identifier in the feature amount database, matches the feature amount data while being encrypted at an arbitrary time point, and performs unique matching based on the matching. A person identifier is paid out for each person. Then, the collation device creates an action database including a set of related information for each person identifier and the photographing device identifier.

  Further, Patent Document 2 proposes a method of dividing feature data extraction processing by two devices.

  In this method, a part of feature data extraction processing for input information is executed by the first information processing apparatus, and the remaining part is executed by the second information processing apparatus. That is, feature data extraction processing is sequentially executed across a plurality of information processing apparatuses. At this time, the first information processing apparatus is characterized in that even if the algorithm is decoded, the final result feature data is not estimated.

  Patent Document 3 proposes a method in which an image acquired by a camera is transmitted to a person tracking device, a head region is detected by the person tracking device, and person information is extracted as a feature amount using the detected position and size. .

  Non-Patent Document 1 proposes a method for detecting a human face region based on gradient information in a video.

  Non-Patent Document 2 describes MPEG-7, which is a standard for expressing feature data.

JP 2011-014059 A JP 2005-222352 A JP 2010-273112 A

Toshinori Hosoi, Tetsuaki Suzuki, Satoshi Sato, “Face Detection by Generalized Learning Vector Quantization”, IEICE Technical Report. PRMU, Pattern Recognition / Media Understanding 102 (651), 47-52, 2003-02-13 Introduction to MPEG-7 Multimedia Content Description Interface, Edited by BS Manjunath, Philippe Salembier, Thomas Sikora, John Wiley & Sons, Ltd., Baffins Lane, Chichester, West Sussex PO19 1UD, England (ISBN 0 471 48678 7) pp.208 ~ Pp.212

  However, just by encrypting the transmission information as in Patent Document 1, if the transmission information is acquired and decrypted by some means, the original image may be generated. In this case, a privacy problem may occur.

  Further, as in Patent Document 3, when a captured image itself is transmitted to an apparatus that extracts a person feature amount in order to reduce the burden on the imaging apparatus, the face of the person is transmitted as it is. This can lead to privacy issues.

  Further, in the technique described in Patent Literature 2, although the algorithm is decoded by the first information processing apparatus, an operation that does not have the possibility of estimating the feature data of the final result is performed. It is. Therefore, if the calculation is performed on all the persons in the image, there is a possibility that the imaging terminal with low processing capability may not be able to process.

  An object of the present invention has been made in view of the above problems, and even if the processing capability of the imaging terminal is low, the person feature amount extraction system that enables the imaging terminal to transmit information protecting privacy to the server Another object is to provide an imaging terminal, a server, and a person feature amount extraction method.

According to the present invention,
At least one imaging terminal and at least one server;
The imaging terminal is
First feature amount extraction means for extracting a first feature amount from an image of a person;
Feature amount transmitting means for transmitting the extracted first feature amount;
The server
Feature quantity receiving means for receiving the transmitted first feature quantity;
Based on the first feature amount, person position specifying means for specifying the position of a person existing in the image;
There is provided a person feature quantity extraction system having second feature quantity extraction means for extracting a second feature quantity for each person based on the first feature quantity and the position of the person.

According to the present invention,
Feature quantity receiving means for receiving the first feature quantity transmitted from the imaging terminal;
Based on the first feature amount, person position specifying means for specifying the position of a person existing in the image;
A server is provided that includes second feature quantity extraction means for extracting a second feature quantity for each person based on the first feature quantity and the position of the person.

According to the present invention,
Using at least one imaging terminal and at least one server,
In the imaging terminal,
First feature amount extraction means for extracting a first feature amount from an image of a person;
Executing a feature amount transmitting means for transmitting the extracted first feature amount;
On the server,
Feature quantity receiving means for receiving the transmitted first feature quantity;
Person position specifying means for specifying the position of a person existing in the image based on the first feature amount;
A person feature amount extraction method is provided that executes second feature amount extraction means for extracting a second feature amount for each person based on the first feature amount and the position of the person.

  According to the present invention, even when the processing capability of the imaging terminal is low, the imaging terminal can transmit the privacy-protected information to the server.

  The above-described object and other objects, features, and advantages will become more apparent from the preferred embodiments described below and the accompanying drawings.

It is a block diagram which shows the structure of the system which concerns on the 1st Embodiment of this invention. It is a flowchart which shows the flow of a process of the system which concerns on the 1st Embodiment of this invention. It is a figure which shows the example of the grid size at the time of extracting the 1st feature value in the 1st Embodiment of this invention. It is a figure which shows the example of creation of a person area. It is a figure which shows the example of the method of extracting an edge histogram feature-value. It is a figure which shows the histogram of the appearance frequency of a brightness | luminance gradient direction. It is a figure which shows the example of the classification method and algorithm of a brightness | luminance gradient direction. It is a figure which shows the example of the method of extracting a color layout feature-value. It is a figure which shows the example of the grid size at the time of extracting the 1st feature value in the 2nd Embodiment of this invention. It is a figure which shows the example of the grid size at the time of extracting the 1st feature value in the 3rd Embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In all the drawings, the same reference numerals are given to the same components, and the description will be omitted as appropriate.

(First embodiment)
FIG. 1 is a block diagram showing a configuration of a system according to the first embodiment of the present invention. The human feature amount extraction system of this embodiment includes an imaging terminal 10 and a server 20. The imaging terminal 10 includes an image acquisition unit 102, a first feature amount extraction unit 104, and a feature amount transmission unit 106. The server 20 includes a feature amount receiving unit 202, a person position specifying unit 204, and a second feature amount extracting unit 206.

  The imaging terminal 10 transmits the first feature amount extracted from the acquired image to the server 20. The server 20 specifies the position of the person based on the first feature quantity received from the imaging terminal 10 and extracts the second feature quantity for identifying the person.

  Note that each component of the imaging terminal 10 and the server 20 illustrated in FIG. 1 is not a hardware unit configuration but a functional unit block. Each component of the imaging terminal 10 and the server 20 includes an arbitrary computer CPU, a memory, a program for realizing the components shown in the figure loaded in the memory, a storage medium such as a hard disk for storing the program, and a network connection interface. It is realized by any combination of hardware and software. There are various modifications of the implementation method and apparatus.

  The processing flow in this embodiment will be described with reference to FIG.

The image acquisition means 102 acquires, for example, an image captured by a video input device such as a CCD digital camera such as a surveillance camera or a video camera, and stores it in a storage area such as a memory or storage (S102). Here, when the information input from the video input device is a moving image, the image acquisition unit 102 cuts out the moving image into arbitrary frames and sets it as an image. Usually, since an image is acquired as an RGB value, the image acquisition unit 102 converts the RGB value of the acquired image into a YCbCr value. The image acquisition unit 102 can convert RGB values into YCbCr values using, for example, the following Equation 1.

  The first feature quantity extraction unit 104 calculates a first feature quantity from the acquired image (S104). In the present embodiment, as an example, the first feature amount extraction unit 104 calculates color distribution information and luminance gradient information as the first feature amount.

  Hereinafter, an example in which the first feature amount extraction unit 104 extracts color distribution information and luminance gradient information will be described.

First, an example in which the first feature amount extraction unit 104 extracts color distribution information will be described. First, as shown in FIG. 3, the first feature quantity extraction unit 104 divides the entire acquired image with a uniform grid size 108 (here, a size of horizontal M × vertical N). The grid size 108 is set in advance in the first feature amount extraction unit 104, for example. Then, the first feature quantity extraction unit 104 is a color (hereinafter referred to as an average color) expressed by an average value of Y, Cb, and Cr of each pixel (m, n) present in each divided grid size 108. Is calculated for each grid using Equation 2 shown below. The first feature quantity extraction unit extracts color distribution information from the acquired image by calculating an average color in each grid of the acquired image using Expression 2.

Next, an example in which the first feature amount extraction unit 104 extracts luminance gradient information will be described. First, when the acquired image is a color image, the first feature amount extraction unit 104 converts the image into a gray scale. Then, the first feature amount extraction unit 104 calculates the strength of the edge gradient in the horizontal direction and the vertical direction from the gray scale image using a Sobel filter or the like. Finally, the first feature quantity extraction unit 104 quantizes the luminance gradient direction in an arbitrary K direction and calculates it along with the gradient strength. Expression 3 shown below is an expression for calculating the gradient strength strength (i, j) in the gray scale (Y component) pixel (i, j). Also, the following Expression 4 is an expression for calculating the edge direction dir (i, j) in the grayscale (Y component) pixel (i, j). The first feature quantity extraction unit 104 calculates the edge direction and the gradient strength at each pixel on the image using the following Expression 3 and Expression 4.

  As described above, the first feature quantity extraction unit 104 only performs a process of averaging the numerical values of pixels or a process of extracting luminance gradient information using a filter. For this reason, the amount of calculation of the processing is smaller than that of arithmetic processing using discrete cosine transform.

  The feature amount transmitting unit 106 transmits the first feature amount extracted from the original image by the first feature amount extracting unit 104 to the server 20 by a signal line such as a LAN or wirelessly such as Wi-Fi (Wireless Fidelity) ( S106). In the present embodiment, the feature amount transmission unit 106 transmits luminance gradient information and color distribution information. When the amount of data to be transmitted is large, the feature amount transmitting unit 106 may compress the data using, for example, run length compression. In the color distribution information extracted by the first feature amount extraction unit 104, information of the same color appears continuously in the divided grid. Therefore, the feature amount transmission unit 106 performs run-length compression, so that the transmission information Efficient compression can be performed. Similarly, in the luminance gradient information extracted by the first feature amount extraction unit 104, the value of the luminance gradient strength for each direction is often 0 in a region other than the edge in the image. For this reason, the feature amount transmitting unit 106 can efficiently compress the transmission information by performing run-length compression on the pixels whose direction-specific luminance gradient strength value is zero. Note that the feature amount transmitting unit 106 may use a compression method other than run-length compression, such as arithmetic coding considering the appearance frequency of each symbol.

  Note that the feature amount transmitting unit 106 transmits the first feature amount without transmitting the original image. Accordingly, even if a third party intercepts transmission information transmitted from the feature amount transmission unit 106, the original image cannot be acquired from the transmission information. Further, the feature amount transmitting unit 106 may not transmit information other than the first feature amount. In this way, privacy is protected with higher accuracy.

  In the server 20, based on the first feature amount transmitted from the feature amount transmitting unit 106, a specific part of the person is detected, the position of the person is estimated, and a second feature amount for specifying the person is extracted. To do. As the specific part of the person, for example, the upper body, the head, the face, and the shoulder can be used. Below, the example at the time of detecting a head or a face as a specific part is explained.

  The feature amount receiving unit 202 receives the first feature amount transmitted from the feature amount transmitting unit 106 and secures it in a storage area such as a memory or a storage. If the received first feature value is compressed by run-length compression or the like, the feature value receiving unit 202 restores the compressed data and secures it in a memory or the like.

  The person position specifying unit 204 detects an area (hereinafter referred to as a person area) where it is estimated that a person is reflected from the first feature amount secured in the memory (S108). In the present embodiment, an example will be described in which the person position specifying unit 204 detects this person area using color distribution information and luminance gradient information.

  First, a method in which the person position specifying unit 204 detects a person area using luminance gradient information will be described.

  The person position specifying means 204 prestores brightness gradient information that serves as a reference for the person's head and face area in a storage unit (not shown). The person position specifying unit 204 compares the stored brightness gradient information with the received brightness gradient information that is the first feature amount, and determines a region showing similarity equal to or higher than a predetermined threshold as the head of the person or Detect as face area. Here, a method for detecting a face region from luminance gradient information quantized in the K direction can be realized by the method described in Non-Patent Document 1. Further, the case of detecting the head can also be realized by using the head image as data for learning in Non-Patent Document 1.

  Then, the person position specifying unit 204 estimates the torso area of the person based on the position and size where the head and face areas are detected. The person position specifying means 204 further includes model creation means (not shown) for creating a model of a person area, and the model creation means is based on the position and size of the head or face area 302 and the upper body area 304 and the lower body area. 306 is determined. FIG. 4 is a diagram illustrating an example of creating a person area. In FIG. 4, an area in which the size of the head or face area 302 is doubled vertically and horizontally is created as an upper body area 304 below the position of the head or face area 302, and below the upper body area 304, As the lower body region 306, an example of a person region in which a region in which the size of the head or face region 302 is expanded twice vertically and horizontally is shown.

  Next, a method in which the person position specifying unit 204 detects a person area using color distribution information will be described.

  The person position specifying means 204 further includes shape storage means (not shown). This shape storage means is a shape of a set of a color indicating a person characteristic (hereinafter referred to as a person color) and a color indicating a characteristic of clothes (hereinafter referred to as a clothes area color) for specifying a person area. Is stored in advance. Then, the person position specifying unit 204 extracts a collective region formed by colors similar to the person color and the clothing region color stored in the shape storage unit from the received color distribution information. Then, the person position specifying unit 204 compares the extracted set area with the set shape stored in the shape storing unit. As a result of the comparison, the person position specifying unit 204 can specify, as a person area, a collective area that shows a degree of similarity greater than a certain threshold. For example, in the case of the head, the person position specifying unit 204 can use the color of the hair as a person color, and can consider the aggregate area formed in a circle with the person color as the head or face area 302. Further, the person position specifying unit 204 can regard the collective area formed in a circle with the person color as the head color or the face area 302 if the face color is a person's face area. Further, the person position specifying unit 204 can regard a collective area formed in a rectangle with the clothes area color as a clothes area color using a specific color such as a uniform as a clothes area. When the head position or face area 302 is extracted, the person position specifying unit 204 can specify the upper body area 304 based on the position and size of the extracted head or face area 302 as in the case of the luminance gradient information. In addition, when the clothes area is extracted, the person position specifying unit 204 may specify the extracted clothes area as the upper body area 304.

  The second feature amount extraction unit 206 calculates a second feature amount representing a pattern or color distribution in the upper body region 304 specified by the person position specifying unit (S110). In the present embodiment, the second feature quantity extraction unit 206 uses a color layout feature quantity representing the color distribution in the upper body area 304 and an edge histogram feature quantity representing a pattern in the upper body area 304 as the second feature quantities, respectively. An example of extraction will be described.

  First, the flow in which the second feature amount extraction unit 206 extracts edge histogram feature amounts based on the luminance gradient information will be described.

  The luminance gradient information is quantized in an arbitrary K direction by the first feature amount extraction unit 104. Therefore, as shown in FIG. 5, the second feature amount extraction unit 206 divides the extracted upper body region 304 into arbitrary M × N grids, and calculates the appearance frequency in the luminance gradient direction of each pixel in each grid. . The appearance frequency of the luminance gradient direction is calculated by classifying the luminance gradient direction in the grid by an angle and the number of appearances for each angle. At this time, pixels whose luminance intensity is below a certain threshold are classified as having no edge. FIG. 6 shows an example in which the appearance frequency is calculated by classifying the luminance gradient directions into four directions A, B, C, and D and E (no edge). In addition, FIG. 7 shows an example of the classification method of the luminance gradient direction and an algorithm for calculating the appearance frequency in each classification method. In FIG. 7, the left and right figures both show examples classified into 8 directions, but the left figure shows the 360 degree direction classified in units of 45 degrees, and the right figure shows the 180 degree direction as 22.2. They are classified in units of 5 degrees. Then, the second feature amount extraction unit 206 divides the calculated appearance frequency by the total number of pixels of the upper body region 304 and performs normalization as the edge histogram feature amount.

  Next, a flow in which the second feature amount extraction unit 206 extracts a color layout feature amount from the color distribution information will be described.

  In the process of extracting the color layout feature quantity, the second feature quantity extraction unit 206 divides the extracted upper body area 304 into 8 × 8 grids, for example, as shown in FIG. 8, and calculates an average color in each grid. . For the extraction of the color layout feature quantity, for example, the method described in pp. 208 to 212 of Non-Patent Document 2 can be used.

  The process of extracting the second feature amount is repeated until the process is completed for all persons specified by the person position specifying unit 204 (S112).

  As described above, according to the present embodiment, the imaging terminal 10 side uses transmission information that cannot identify an individual, and does not perform a calculation with a large calculation amount in the processing. Therefore, even in a system that requires the server 20 that has a low processing capability of the imaging terminal 10 and is capable of advanced processing, it is possible to perform personal identification that protects privacy.

(Second Embodiment)
This embodiment is the same as the first embodiment except for the following points.

  FIG. 9 is a diagram illustrating an example of the grid size 108 when extracting the first feature amount in the second embodiment of the present invention. In the present embodiment, the grid size 108 is set so that the grid increases from the top of the image to the bottom of the image.

  According to this configuration, the grid size 108 for the person on the near side that appears larger on the image is set to be larger. Therefore, compared with the case where the average color is calculated with the grid size 108 having a uniform size, the image of the person appearing large is coarse. For this reason, even if someone visually recognizes the information transmitted by the feature amount transmission unit 106 to the server, it becomes more difficult to specify a person who appears to be large. Furthermore, since the area where the same color information continues is increased by increasing the grid size 108, the feature amount transmission unit 106 performs run length compression or the like as compared with the case where the average color is calculated with the uniform grid size 108. When is used, the compression efficiency of the image can be increased.

  As described above, also in this embodiment, the same effect as that of the first embodiment can be obtained. In the present embodiment, the grid size 108 increases from the upper part of the image to the lower part of the image, so that even a person who appears large in the image is identified as an individual as compared to a case where the grid size 108 is divided. It becomes difficult to protect privacy more firmly. Furthermore, the image compression efficiency when the feature amount transmission means 106 uses run length compression or the like is improved, and the communication cost can be reduced.

(Third embodiment)
This embodiment is the same as the second embodiment except for the following points.

  FIG. 10 is a diagram illustrating an example of the grid size 108 when extracting the first feature amount in the third embodiment of the present invention. In the present embodiment, the imaging terminal 10 holds three-dimensional position information in an image calculated based on camera calibration information. The first feature quantity extraction unit 104 estimates the size (hereinafter referred to as an assumed person size) in which a person appears according to the position on the image based on the stored three-dimensional position information. The first feature quantity extraction unit 104 determines the grid size 108 based on the estimated assumed person size.

  Here, the assumed person size is estimated based on a reference (hereinafter referred to as a reference person size) such as an average person model based on average height or each average model based on information collected for each age and sex. can do. Further, a plurality of these models can be adopted as the reference person size. The first feature quantity extraction unit 104 estimates the size of the person appearing at an arbitrary position on the image by expanding and contracting the reference person size against the three-dimensional position information in the image. be able to. At this time, if the size of the grid size 108 is associated with the reference person size in advance, the first feature amount extraction unit 104 can dynamically expand and contract the grid size 108 based on the above estimation.

  As described above, also in this embodiment, the same effects as those in the first and second embodiments can be obtained. In the present embodiment, since the imaging terminal 10 further has the three-dimensional position information and the assumed person size, the grid size 108 can be dynamically changed according to the size in which the person is reflected. Therefore, the optimum grid size 108 can be automatically selected.

In addition, according to embodiment mentioned above, the following invention is disclosed.
(Appendix 1)
At least one imaging terminal and at least one server;
The imaging terminal is
First feature amount extraction means for extracting a first feature amount from an image of a person;
Feature amount transmitting means for transmitting the extracted first feature amount;
The server
Feature quantity receiving means for receiving the transmitted first feature quantity;
Based on the first feature amount, person position specifying means for specifying the position of a person existing in the image;
A person feature quantity extraction system comprising second feature quantity extraction means for extracting a second feature quantity for each person based on the first feature quantity and the position of the person.
(Appendix 2)
In the human feature amount extraction system according to attachment 1,
The imaging terminal is a person feature extraction system that does not transmit the image.
(Appendix 3)
In the person feature extraction system according to appendix 1 or 2,
The person feature amount extraction system, wherein the first feature amount includes at least one of color distribution information and luminance gradient information.
(Appendix 4)
In the person feature extraction system according to any one of appendices 1 to 3,
The first feature amount extraction unit divides the image using a variable grid size whose vertical and horizontal sizes increase from the upper part of the image toward the lower part of the image, and uses the feature amount of each divided image, A person feature extraction system for extracting the first feature.
(Appendix 5)
In the person feature extraction system according to appendix 4,
The imaging terminal is
Holding 3D position information indicating a 3D position on the image;
The first feature amount extraction means includes:
Based on the three-dimensional position information, an assumed person size that is an assumed size that a person appears on the image is calculated,
A person feature extraction system that determines the grid size based on the assumed person size.
(Appendix 6)
In the person feature extraction system according to appendix 5,
The assumed person size is a person feature amount extraction system calculated based on a reference person size including an average person model based on average height or an average person model for each gender and age.
(Appendix 7)
In the person feature extraction system according to any one of appendices 4 to 6,
The first feature amount has color distribution information,
The person feature quantity extraction system in which the feature quantity for each of the divided images based on the color distribution information is an average color calculated from colors existing in the divided image.
(Appendix 8)
First feature amount extraction means for extracting a first feature amount from an image of a person;
An imaging terminal having a feature amount transmitting means for transmitting the extracted first feature amount without transmitting the image.
(Appendix 9)
Feature quantity receiving means for receiving the first feature quantity transmitted from the imaging terminal;
Person position specifying means for specifying the position of a person existing in the image based on the first feature amount;
A server having second feature quantity extraction means for extracting a second feature quantity for each person based on the first feature quantity and the position of the person.
(Appendix 10)
Using at least one imaging terminal and at least one server,
In the imaging terminal,
A first feature extraction process for extracting a first feature from an image of a person;
Executing a feature amount transmission process for transmitting the extracted first feature amount;
On the server,
A feature amount receiving process for receiving the transmitted first feature amount;
A person position specifying process for specifying a position of a person existing in the image based on the first feature amount;
A person feature quantity extraction method for executing a second feature quantity extraction process for extracting a second feature quantity for each person based on the first feature quantity and the position of the person.
(Appendix 11)
In the imaging terminal according to attachment 8,
The imaging terminal, wherein the first feature amount includes at least one of color distribution information and luminance gradient information.
(Appendix 12)
In the imaging terminal according to appendix 8 or 11,
The first feature amount extraction unit divides the image using a variable grid size whose vertical and horizontal sizes increase from the upper part of the image toward the lower part of the image, and uses the feature amount of each divided image, An imaging terminal for extracting the first feature amount.
(Appendix 13)
In the imaging terminal according to attachment 12,
The imaging terminal is
Holding 3D position information indicating a 3D position on the image;
The first feature amount extraction means includes:
Based on the three-dimensional position information, an assumed person size that is an assumed size that a person appears on the image is calculated,
An imaging terminal that determines the grid size based on the assumed person size.
(Appendix 14)
In the imaging terminal according to attachment 13,
The assumed terminal size is an imaging terminal that is calculated based on a reference person size including an average person model based on average height or an average person model for each gender and age.
(Appendix 15)
In the imaging terminal according to any one of appendices 12 to 14,
The first feature amount has color distribution information,
The imaging terminal, wherein the feature amount for each divided image based on the color distribution information is an average color calculated from colors existing in the divided image.
(Appendix 16)
In the person feature amount extraction method according to attachment 10,
A person feature extraction method in which the imaging terminal does not transmit the image.
(Appendix 17)
In the person feature extraction method according to appendix 10 or 16,
The person feature amount extraction method, wherein the first feature amount includes at least one of color distribution information and luminance gradient information.
(Appendix 18)
In the person feature extraction method according to any one of Supplementary Notes 10 or 16 or 17,
In the first feature amount extraction process, the image is divided using a variable grid size in which the vertical and horizontal sizes increase from the upper part of the image toward the lower part of the image, and the feature amount for each divided image is used. A person feature extraction method for extracting the first feature.
(Appendix 19)
In the person feature amount extraction method according to attachment 18,
The imaging terminal is
Holding 3D position information indicating a 3D position on the image;
The first feature amount extraction process includes:
Based on the three-dimensional position information, an assumed person size that is an assumed size that a person appears on the image is calculated,
A person feature extraction method for determining the grid size based on the assumed person size.
(Appendix 20)
In the person feature amount extraction method according to attachment 19,
The assumed person size is a person feature extraction method calculated based on a reference person size including an average person model based on average height or an average person model for each gender and age.
(Appendix 21)
In the person feature extraction method according to any one of appendices 18 to 20,
The first feature amount has color distribution information,
The person feature amount extraction method, wherein the feature amount for each of the divided images based on the color distribution information is an average color calculated from colors existing in the divided image.

  As mentioned above, although embodiment of this invention was described with reference to drawings, these are the illustrations of this invention, Various structures other than the above are also employable. Each embodiment has a configuration of one imaging terminal 10 and one server 20 for convenience of explanation, but a configuration in which a plurality of imaging terminals 10 or servers 20 exist may be employed.

  This application claims the priority on the basis of Japanese application Japanese Patent Application No. 2012-000388 for which it applied on January 5, 2012, and takes in those the indications of all here.

Claims (9)

At least one imaging terminal and at least one server;
The imaging terminal is
First feature amount extraction means for extracting a first feature amount from an image of a person;
Feature amount transmitting means for transmitting the extracted first feature amount;
The server
Feature quantity receiving means for receiving the transmitted first feature quantity;
Based on the first feature amount, person position specifying means for specifying the position of a person existing in the image;
A person feature quantity extraction system comprising second feature quantity extraction means for extracting a second feature quantity for each person based on the first feature quantity and the position of the person. The person feature amount extraction system according to claim 1,
The imaging terminal is a person feature extraction system that does not transmit the image. The person feature amount extraction system according to claim 1 or 2,
The person feature amount extraction system, wherein the first feature amount includes at least one of color distribution information and luminance gradient information. The person feature amount extraction system according to any one of claims 1 to 3,
The first feature amount extraction unit divides the image using a variable grid size whose vertical and horizontal sizes increase from the upper part of the image toward the lower part of the image, and uses the feature amount of each divided image, A person feature extraction system for extracting the first feature. The person feature amount extraction system according to claim 4,
The imaging terminal is
Holding 3D position information indicating a 3D position on the image;
The first feature amount extraction means includes:
Based on the three-dimensional position information, an assumed person size that is an assumed size that a person appears on the image is calculated,
A person feature extraction system that determines the grid size based on the assumed person size. The person feature amount extraction system according to claim 5,
The assumed person size is a person feature amount extraction system calculated based on a reference person size including an average person model based on average height or an average person model for each gender and age. The person feature amount extraction system according to any one of claims 4 to 6,
The first feature amount has color distribution information,
The person feature quantity extraction system in which the feature quantity for each of the divided images based on the color distribution information is an average color calculated from colors existing in the divided image. Feature quantity receiving means for receiving the first feature quantity transmitted from the imaging terminal;
Person position specifying means for specifying the position of a person existing in the image based on the first feature amount;
A server having second feature quantity extraction means for extracting a second feature quantity for each person based on the first feature quantity and the position of the person. Using at least one imaging terminal and at least one server,
In the imaging terminal,
A first feature extraction process for extracting a first feature from an image of a person;
Executing a feature amount transmission process for transmitting the extracted first feature amount;
On the server,
A feature amount receiving process for receiving the transmitted first feature amount;
A person position specifying process for specifying a position of a person existing in the image based on the first feature amount;
A person feature quantity extraction method for executing a second feature quantity extraction process for extracting a second feature quantity for each person based on the first feature quantity and the position of the person.

Images