JP6424309B1 - Program and apparatus for generating a three-dimensional model based on measurement values - Google Patents

Abstract

Provided is a program and an apparatus capable of easily generating a three-dimensional model from only measured values and encoding and decoding to a very small data capacity.
A three-dimensional model is generated from a measurement value of a dimension number n of one body. The apparatus outputs, for each 3D model, a component variable having a dimension number m that is dimension-compressed using a teacher data group in which measurement values having a dimension number n corresponding to a plurality of measurement locations are associated with each other. A statistical learning engine that constructs a learning model, a correlation learning engine that constructs a correlation learning model of a dimension value of dimension n and a component variable of dimension number m from a plurality of three-dimensional models of a teacher data group, and a correlation Using a learning engine, an encoding means for encoding from a measurement value of one dimension number n as target data to a component variable of dimension number m, and a three-dimensional from the component variable of dimension number m using a statistical learning engine Decoding means for decoding into a model.
[Selection] Figure 1

Description

  The present invention relates to a technique for generating a three-dimensional model. Specifically, it relates to encoding and decoding technology based on a three-dimensional model.

  In recent years, there is a technique of a three-dimensional scanner capable of detecting human body shape data (see, for example, Non-Patent Documents 1 and 2). According to this technique, three-dimensional point cloud data is measured by non-contact optical triangulation with respect to the human body. These point cloud data are extremely high density of about 1 million points, and realize extremely small errors in human body measurement applications. Such human body shape data is also effective as health management data other than body weight.

  Conventionally, there is a technique for deforming a muscle model superimposed on a skeleton model in accordance with a measurement result of a subject (see, for example, Patent Document 1). According to this technique, a human body model corresponding to the subject himself / herself is created based on the physical measurement result of the subject by the body composition meter and the three-dimensional measuring device. The human body model is an anatomical model in which a skeleton, muscles and fat are set, and these are deformed according to the measurement result of the subject. By switching and displaying these models of skeleton, muscle, and fat, the subject can visually recognize the situation inside the body.

  There is also a technique in which a three-dimensional model expressing an object with feature parameters is stored in advance, and an image of a feature region detected from a captured image is adapted to the three-dimensional model (see, for example, Patent Document 2). According to this technique, the value of the feature parameter of the three-dimensional model representing the object imaged in the feature region image is calculated. Then, the data amount of the entire three-dimensional image is reduced by outputting the value of the feature parameter and the image of the region other than the feature region.

JP 2017-176803 A JP 2009-268088 A

“3D BODY SCANNER SCUVEG4”, Space Vision, Inc., [online], [searched July 14, 2018], Internet <URL: http://spacevision.ap-northeast-1.elasticbeanstalk.com/productservice/3d -body-scanner-scuveg4 / ＞ “3D Body Station”, 3D Body Lab, Inc., [online], [searched July 14, 2018], Internet <URL: https://www.3dbodylab.co.jp/3dbodystation/>

  In the case of Non-Patent Documents 1 and 2 described above, it is necessary to use a three-dimensional scanner that is large in scale and cost in order to generate a three-dimensional model of a human body. In order to improve the accuracy of the 3D model, the number of vertices is set to 15,000 or more, for example, and each vertex is expressed in 3 dimensions (x, y, z). The huge amount of data.

On the other hand, the inventor of the present application thought that a three-dimensional model close to the user's body shape could be easily generated from the measured values without preparing a three-dimensional scanner for optical triangulation. For example, there is a technology for creating a three-dimensional model by photographing one's body shape with a smartphone camera, but it is only for the purpose of generating an agent character such as an avatar, and the accuracy of the three-dimensional model is extremely low .
On the other hand, when the user selects the size of the clothes on a daily basis, the size is selected based only on the measurement values of some of the measurement locations. That is, the measurement value of the measurement location is a reference representing the user's body shape.

  Further, the inventor of the present application thought that the data of the three-dimensional model could be easily shared (transmitted / received) with a small capacity. In general, in the clothing industry, for example, it is possible to select clothes and adjust measurement from a three-dimensional model representing a user's body shape. Further, for example, in the medical industry, there may be an application that represents a three-dimensional model of each organ from DICOM (Digital Imaging and Communication in Medicine) data. Furthermore, in the advertising industry, an application in which an agent character adapted to the body shape of the user is displayed on a signage display is also conceivable. Therefore, it is necessary to instantaneously transmit and receive the three-dimensional model data between the user and the service provider.

  Furthermore, the inventor of the present application thought that the three-dimensional model expressing the user's own body shape should be kept confidential as personal information for the user. That is, it is preferable that the data itself that encodes the three-dimensional model is encrypted so that it cannot be understood by a third party.

  Therefore, an object of the present invention is to provide a program and an apparatus that can easily generate a three-dimensional model from only the measurement values and encode and decode it to an extremely small data capacity.

According to the present invention, there is provided a program for causing a computer mounted on an apparatus for generating a three-dimensional model composed of a plurality of vertices to function from a measurement value of one dimension number n as target data,
As the teacher data group, for each three-dimensional model, dimension values of n dimensions corresponding to a plurality of measurement locations are associated,
A statistical learning engine that outputs a component variable of dimension number m from a plurality of three-dimensional models of a teacher data group and constructs a statistical learning model;
A correlation learning engine that constructs a correlation learning model between a dimension n measurement value and a dimension m component variable from a plurality of three-dimensional models of a teacher data group;
Encoding means for encoding from a measurement value of one dimension number n as target data to a component variable of dimension number m using a correlation learning engine;
Using a statistical learning engine, the computer is caused to function as decoding means for decoding a component variable having the dimension number m into a three-dimensional model.

According to the present invention, there is provided a program for causing a computer mounted on an apparatus for encoding a three-dimensional model composed of a plurality of vertices to function.
As the teacher data group, for each three-dimensional model, dimension values of n dimensions corresponding to a plurality of measurement locations are associated,
A statistical learning engine that outputs a component variable of dimension number m from a plurality of three-dimensional models of a teacher data group and constructs a statistical learning model;
A correlation learning engine that constructs a correlation learning model between a dimension n measurement value and a dimension m component variable from a plurality of three-dimensional models of a teacher data group;
Using the correlation learning engine, the computer is caused to function as an encoding means for encoding from a measuring value of one dimension number n as target data to a component variable of dimension number m.

According to another embodiment of the program of the present invention,
In order to extract the measurement value of dimension number n corresponding to a plurality of measurement locations from the 3D model itself of the teacher data group,
Using a statistical learning engine, a component variable having a dimension number m is output by inputting a three-dimensional model of the teacher data group, and reversely reproduced by inputting the component variable having the dimension number m output in reverse reproduction. It is also preferable to output the generated three-dimensional model and cause the computer to function so as to extract the measurement value of the number of dimensions n corresponding to a plurality of measurement locations from the reverse reproduction three-dimensional model itself.

According to another embodiment of the program of the present invention,
The encoding means is
As the target data, only k (<n) measurement values are determined for the measurement value of the dimension number n of one body, and the component variable of the dimension number m is determined even if the other n−k measurement values are missing. To estimate
It is also preferable to make the computer function so as to calculate the component variable having the dimension number m including the other n−k measurement values that are optimized using the k measurement values as a constraint.

According to another embodiment of the program of the present invention,
The encoding means also preferably causes the computer to function so as to use Lagrange's method of Lagrange multiplier.

According to another embodiment of the program of the present invention,
It is also preferable to further cause the computer to function as tag creation means for creating a tag that describes the encoded component variable having the dimension number m.

According to another embodiment of the program of the present invention,
A teacher data group is prepared for each of the three-dimensional models of different model types,
A plurality of statistical learning engines and correlation learning engines are provided with different teacher data groups for each model type,
Encoding means uses a correlation learning engine that corresponds to a model based identifiers measurement value of the number of dimensions n, encodes the measuring value of the number of dimensions n to component variable dimensionality m,
The tag creating means preferably causes the computer to function so that the model type identifier and the component variable having the dimension number m based on the model type identifier are associated with each other and described in the tag.

According to the present invention, there is provided a program for causing a computer mounted on an apparatus for decoding a three-dimensional model composed of a plurality of vertices to function.
As the teacher data group, for each three-dimensional model, dimension values of n dimensions corresponding to a plurality of measurement locations are associated,
A statistical learning engine that outputs a component variable of dimension number m from a plurality of three-dimensional models of a teacher data group and constructs a statistical learning model;
A correlation learning engine that constructs a correlation learning model between a dimension n measurement value and a dimension m component variable from a plurality of three-dimensional models of a teacher data group;
Using a statistical learning engine, the computer is caused to function as decoding means for decoding a component variable having a dimension number m as target data into a three-dimensional model.

According to another embodiment of the program of the present invention,
It is also preferable that the decoding means causes the computer to function to derive a dimension value of dimension number n from a component variable of dimension number m as target data using a correlation learning engine.

According to another embodiment of the program of the present invention,
It is also preferable to cause the computer to function as a tag reading unit that reads a component variable having a dimension number m from a tag in which a component variable having a dimension number m is described.

According to another embodiment of the program of the present invention,
A teacher data group is prepared for each of the three-dimensional models of different model types,
A plurality of statistical learning engines and correlation learning engines are provided with different teacher data for each model type,
The tag reading means reads the model type identifier and the component variable having the dimension number m based on the model type identifier from the tag,
The decoding means preferably uses a correlation learning engine corresponding to the read model type identifier to decode the component variable having the number of dimensions m into a three-dimensional model to cause the computer to function.

According to another embodiment of the program of the present invention,
It is also preferable that the tag causes the computer to function as a QR (Quick Response) (registered trademark) code or an RFID (Radio Frequency IDentifier).

According to another embodiment of the program of the present invention,
The three-dimensional model is shape data of a human body or an object, and is expressed with the same number of vertices with respect to the same object.
It is also preferable to make the computer function so that the number of plural teacher data groups is smaller than the number of vertices of the three-dimensional model.

According to another embodiment of the program of the present invention,
The statistical learning engine is based on Principal Component Analysis,
It is also preferred to have the computer function so that the correlation learning engine is based on the least squares method.

According to the present invention, an apparatus for generating a three-dimensional model composed of a plurality of vertices from a measurement value of a dimension number n as a target data,
As the teacher data group, for each three-dimensional model, dimension values of n dimensions corresponding to a plurality of measurement locations are associated,
A statistical learning engine that outputs a component variable of dimension number m from a plurality of three-dimensional models of a teacher data group and constructs a statistical learning model;
A correlation learning engine that constructs a correlation learning model between a dimension n measurement value and a dimension m component variable from a plurality of three-dimensional models of a teacher data group;
Encoding means for encoding from a measurement value of one dimension number n as target data to a component variable of dimension number m using a correlation learning engine;
And a decoding means for decoding the component variable having the dimension number m into a three-dimensional model using a statistical learning engine.

According to the present invention, an apparatus for encoding a three-dimensional model consisting of a plurality of vertices,
As the teacher data group, for each three-dimensional model, dimension values of n dimensions corresponding to a plurality of measurement locations are associated,
A statistical learning engine that outputs a component variable of dimension number m from a plurality of three-dimensional models of a teacher data group and constructs a statistical learning model;
A correlation learning engine that constructs a correlation learning model between a dimension n measurement value and a dimension m component variable from a plurality of three-dimensional models of a teacher data group;
It has an encoding means for inputting a measuring value of one dimension number n as target data and encoding it into a component variable of dimension number m using a correlation learning engine.

According to the present invention, an apparatus for decoding a three-dimensional model composed of a plurality of vertices,
As the teacher data group, for each three-dimensional model, dimension values of n dimensions corresponding to a plurality of measurement locations are associated,
A statistical learning engine that outputs a component variable of dimension number m from a plurality of three-dimensional models of a teacher data group and constructs a statistical learning model;
A correlation learning engine that constructs a correlation learning model between a dimension n measurement value and a dimension m component variable from a plurality of three-dimensional models of a teacher data group;
Decoding means for decoding from a component variable of dimension number m as target data into a three-dimensional model using a statistical learning engine.

  According to the program and apparatus of the present invention, it is possible to easily generate a three-dimensional model from only the measured values and to encode and decode it to an extremely small data capacity.

It is a functional block diagram of the apparatus which produces | generates the three-dimensional model in this invention. It is explanatory drawing showing the vector space of the three-dimensional model in a statistical learning engine. It is explanatory drawing showing the statistical shape space in a statistical learning engine. A simple code representing principal component analysis in a statistical learning engine. It is explanatory drawing showing the measurement value space derived | led-out from the measurement value of the measurement location. It is explanatory drawing showing the linear transformation of a statistical shape space and a measurement value space. It is a simple code representing a linear transformation between a statistical shape space and a measurement value space in a correlation learning engine. It is explanatory drawing showing the precision of the three-dimensional model decoded by this invention. It is explanatory drawing showing the influence which the change of the number of measurement places in this invention has on a three-dimensional model. It is a functional block diagram of the apparatus which encodes and decodes a three-dimensional model to a tag. It is a functional block diagram which reconfigure | reconstructs matching with a three-dimensional model and a measurement value. It is a functional block diagram which switches a learning engine according to a model classification. It is explanatory drawing showing the measurement value space which estimates the missing value in this invention. It is a simple code | symbol showing the missing value estimation in this invention. It is a user interface showing the three-dimensional model, measurement value, and QR code in the present invention.

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

  FIG. 1 is a functional configuration diagram of an apparatus for generating a three-dimensional model in the present invention.

1 is a functional configuration diagram of an apparatus for generating a three-dimensional model by inputting a plurality of measurement values.
According to FIG. 1, the apparatus 1 includes a statistical learning engine 101, a correlation learning engine 102, an encoding unit 111, and a decoding unit 122. These functional components can be realized by executing a program that causes a computer installed in the apparatus to function.

[Statistical learning engine 101]
The statistical learning engine 101 inputs a plurality of three-dimensional models of the teacher data group, outputs dimension-compressed component variables of the number m of dimensions, and constructs a statistical learning model.

  FIG. 2 is an explanatory diagram showing a vector space of a three-dimensional model in the statistical learning engine.

According to FIG. 2, for example, 1,000 human bodies having various body shapes are assumed as the teacher data. The three-dimensional model is shape data of a human body or an object, and is expressed with the same number of vertices with respect to the same object.
According to FIG. 2A, the three-dimensional model has N = 15,000 vertices for each body, and each vertex is expressed in three dimensions (x, y, z). That is, one 3D model is represented by a 3N (= 45,000) dimensional vector.
According to FIG. 2B, each 3D model is represented by one point in the 3N-dimensional space.

  According to the general machine learning engine, an enormous number of teacher data is required, whereas according to the present invention, the number of teacher data groups is smaller than the number of vertices of the three-dimensional model. May be. That is, the number of human bodies of teacher data is 1,000, which is smaller than the number of vector dimensions of 45,000 in the three-dimensional model. That is, it is not necessary to prepare the number of teacher data pieces more than the number of vector dimensions of the three-dimensional model.

FIG. 3 is an explanatory diagram showing a statistical shape space in the statistical learning engine.
FIG. 4 is a simple code representing principal component analysis in the statistical learning engine.

Specifically, the statistical learning engine 101 is based on Principal Component Analysis.
By “principal component analysis”, a small number (for example, 30) principal components (component variables) that are uncorrelated with each other and best represent the overall variation are derived from 1000 correlated 3N-dimensional spaces. The variance of the first principal component is maximized, and the subsequent principal components are selected so as to maximize the variance under the constraint condition that is uncorrelated with the principal components determined so far. By maximizing the variance of the principal component, the principal component has as much explanatory ability as possible with respect to changes in the observed value. The principal axis giving the principal component is an orthogonal basis of a group of 1000 points in 3N-order space. The orthogonality of the principal axes is derived from the fact that the principal axes are eigenvectors of the covariance matrix and the covariance matrix is a real symmetric matrix.

The statistical learning engine 101 constructs a statistical learning model that projects a 3N-dimensional space into a statistical shape space (for example, 30 dimensions) having the number of component variables based on principal component analysis as the number of dimensions.
According to the present invention, each 3D model in 3N (= 45,000) dimensional space is projected onto, for example, 30 dimensional (component variable) space. The transformation that gives the principal component is represented by a singular value decomposition of a matrix made up of a set of observation values, and the singular value decomposition of a rectangular matrix X made up of a group of 1000 points in a 3N-dimensional space is represented by the following equation.
X = U * Σ * V ^T
X: Matrix of 1000 points in 3N dimensional space (1000 rows x 3N columns)
U: square matrix of n (1000) x n (1000) (orthogonal matrix of n-dimensional unit vectors)
Σ: rectangular diagonal matrix of n (1000) x p (3N) (diagonal component is the singular value of X)
V: p (3N) × p (3N) square matrix (p-dimensional unit vector orthogonal matrix)
Here, the matrix of the first 30 columns of V is changed to V. The linear transformation by the matrix V gives the principal component of X.
V: Matrix representing transformation to 3N dimensional space-> statistical shape (30 dimensional) space
V ^-1 : Matrix representing transformation to statistical shape (30 dimensions) space-> 3N dimensional space Note that the superscript -1 of the matrix is not a symbol indicating an inverse matrix, but for the transformation of the vector defined by the matrix It is used as an abstract symbol that means the inverse transformation. Here, V ⁻¹ is equal to the transpose V ^T of V.

As is clear from FIG. 3, the three-dimensional model can be associated with each other between the 3N-dimensional space and the statistical shape space by linear transformation using the matrix V or V ⁻¹ .
s = x * V
x = s * V ⁻¹
s: Statistical shape space vector
x: vector in 3N-dimensional space
V: Statistical learning model

[Correlation learning engine 102]
The correlation learning engine 102 constructs a correlation learning model of a dimension n measurement value and a component variable m dimension from a plurality of three-dimensional models of the teacher data group.

  FIG. 5 is an explanatory diagram showing a measurement value space derived from the measurement values at the measurement location.

As a teacher data group, for each three-dimensional model, a dimension value n corresponding to a plurality of measurement locations is associated. According to FIG. 5, the measurement values of a plurality of measurement locations are assigned to the human body of the three-dimensional model of the teacher data in FIG. 2 described above. In this case, it is possible to derive a measurement value space having the measurement location as an element and the measurement value as the element value. For example, when the measurement values of 10 measurement locations are given, the measurement value space is 10 dimensions.
Of course, the measurement location may be one or more. It may be height only, height + waist circumference, or height + abdominal circumference + chest circumference.

  Here, the data of the 3D model and the measurement value of the measurement location may be associated with each other separately, or the measurement value of the measurement location can be extracted from the 3D model itself. May be.

FIG. 6 is an explanatory diagram showing linear transformation between the statistical shape space and the measurement value space.
FIG. 7 is a simple code representing a linear conversion between the statistical shape space and the measurement value space in the correlation learning engine.

The correlation learning engine 102 is based on the least square method. The “least squares method” is a method of calculating the residual squared when approximating a linear model from a plurality of multidimensional vectors (data sets). Determining the most probable linear model with the smallest sum.

As is clear from FIG. 5, the three-dimensional model can be associated with each other between the statistical shape space and the measurement value space by linear transformation using the matrix A or A ⁻¹ .
s = d * A
d = s * A ⁻¹
A = (D ^T * D) ⁻¹ * D ^{T *} S (Deriving A that minimizes || D * A−S ||)
s: Statistical shape space vector
d: Vector of measuring value space
S: Vector set of statistical shape space
D: Vector set of measuring value space
A: Correlation learning model

[Encoding unit 111]
Using the correlation learning engine 102, the encoding unit 111 encodes a single dimension number n measurement value as target data into a dimension number m component variable. The encoded component variable of dimension number m has confidentiality that cannot recognize the measurement value. Therefore, it is suitable when the measurement value of the dimension number n of one object data is confidential information.

[Decoding unit 122]
The decoding unit 122 uses the statistical learning engine 101 to decode the encoded component variable having the number of dimensions m into a three-dimensional model.
Of course, the dimension value of dimension n can be extracted from the decoded three-dimensional model in a straight-line manner.

  FIG. 8 is an explanatory diagram showing the accuracy of the three-dimensional model decoded by the present invention.

FIG. 8A shows four different three-dimensional models A to D as target data.
FIG. 8B shows a three-dimensional model in which each of the three-dimensional models A to D of the target data is encoded into component variables by the statistical learning engine 101 and decoded by the statistical learning engine 101 from the component variables. Comparing FIG. 8A and FIG. 8B, it can be understood that the three-dimensional model is maintained in substantially the same shape even if encoding and decoding are performed by principal component analysis.
FIG. 8C shows the three-dimensional models A to D of the target data that are encoded into component variables from the measurement values (10 locations) by the correlation learning engine 102 and decoded by the statistical learning engine 101 from the component variables. Represents a model. Comparing FIG. 8 (a) and FIG. 8 (c), it can be understood that the three-dimensional model is maintained in substantially the same shape even if encoding and decoding are performed according to the measurement value.

  FIG. 9 is an explanatory diagram showing the influence of the change in the number of measurement locations in the present invention on the three-dimensional model.

FIG. 9A shows three different three-dimensional models A to C as target data.
FIG. 9B shows the three-dimensional models A to C of the target data, which are encoded as a measurement value from “height only” into a component variable by the correlation learning engine 102, and decoded from the component variable by the statistical learning engine 101. Represents a dimensional model. Comparing FIG. 9A and FIG. 9B, the identity of the shape of the three-dimensional model cannot be recognized except for the height.
In FIG. 9C, for each of the three-dimensional models A to C of the target data, the measurement values are encoded from the “height” and “waist circumference” into the component variables by the correlation learning engine 102, and decoded from the component variables by the statistical learning engine 101. Represents a three-dimensional model. By comparing FIG. 9A and FIG. 9C, the identity of the shape of the three-dimensional model can be recognized more than in FIG. 9B.
FIG. 9 (d) shows that the three-dimensional models A to C of the target data are encoded as component values by the correlation learning engine 102 from "height", "abdominal circumference", and "chest circumference" as measurement values, and the statistical learning engine from the component variables. The three-dimensional model decoded by 101 is represented. By comparing FIG. 9A and FIG. 9D, the identity of the shape of the three-dimensional model can be recognized.
In other words, only the three-dimensional measurement values of “height”, “abdominal circumference”, and “chest circumference” can be decoded into a shape almost similar to the original three-dimensional model.

  FIG. 10 is a functional configuration diagram of an apparatus that encodes and decodes a three-dimensional model into a tag.

10, compared with FIG. 1, it has the tag production | generation part 112 as an encoding, and the tag reading part 121 as a decoding.
[Tag creation unit 112]
The tag creation unit 112 inputs the component variable having the dimension number m output from the encoding unit 111 and creates a tag describing the component variable.
[Tag reading unit 121]
The tag reading unit 121 reads the encoded component variable having the dimension number m from the tag. The read component variable is output to the decoding unit 122.

The tag may be a QR (Quick Response) code, for example. The QR code is a matrix type two-dimensional code, and can describe a maximum of 2,953 bytes in binary. In a general smartphone, the QR code can be displayed on a display, and the QR code can be read by a camera.
According to the present invention, when a three-dimensional model is 30 dimensions with component variables (4 bytes), it can be represented with 120 bytes. If 120 bytes can be described in the QR code, for example, a three-dimensional model representing the user's own body shape can be clearly indicated by the QR code.

The tag is not limited to a QR code, but may be an RFID (Radio Frequency IDentifier). RFID refers to a technology for communicating information described in an RF tag by short-range wireless communication using an electromagnetic field or a radio wave. For example, it is Felica (registered trademark) and is used for electronic money and a boarding card.
According to the present invention, for example, a component variable of a three-dimensional model can be instantaneously read by a reader simply by describing it in an RF tag. The reader that reads the component variable from the RF tag can instantaneously display the three-dimensional model corresponding to the component variable on the display.

  FIG. 11 is a functional configuration diagram for reconfiguring the association between the three-dimensional model and the measurement value.

According to FIG. 11, the measurement value of dimension number n corresponding to a plurality of measurement locations is extracted from the three-dimensional model itself of the teacher data group. That is, it is assumed that the measurement value is extracted from the three-dimensional model itself without associating the measurement value separately from the three-dimensional model.
For this purpose, the statistical learning engine 101 inputs a three-dimensional model of teacher data and outputs a component variable having a dimension number m. Here, in reverse reproduction, the statistical learning engine 101 outputs a reverse reproduction three-dimensional model by inputting the outputted component variable having the dimension number m. From the reversely reproduced three-dimensional model itself, a dimension n measurement value corresponding to a plurality of measurement locations is extracted.

Since the statistical learning engine 101 performs dimensional compression by principal component analysis, the three-dimensional model of the teacher data and the three-dimensional model reproduced from the component variable having the dimension number m include an error. Therefore, if the measurement value is extracted from the reverse reproduction three-dimensional model itself decoded from the component variable having the dimension number m, the component variable having the dimension number m and the measurement value can be associated with high accuracy.
The correlation learning engine 102 learns by associating the component variable of the number of dimensions m output from the statistical learning engine 101 with the measurement value of the number of dimensions n extracted from the reverse reproduction three-dimensional model itself.

  FIG. 12 is a functional configuration diagram for switching the learning engine in accordance with the model type.

According to FIG. 12, a teacher data group is prepared for each of the three-dimensional models of different model types. According to the above-described embodiment, the three-dimensional model is described as a human body, but various objects may be used.
In the case of medical use, for example, a 3D model for each organ can be created from DICOM data of a medical image taken by MRI (nuclear magnetic resonance imaging) or CT (computer tomography). For example, by assigning a model type to each organ, the three-dimensional model can be switched according to the model type.

  A teacher data group is prepared for each of the three-dimensional models of different model types. A plurality of statistical learning engines 101 and correlation learning engines 102 are provided with different teacher data groups for each model type.

As the encoding side, the encoding unit 111 inputs a dimension value of dimension number n in the target data and a model ID (model type identifier) of the target data. Then, the encoding unit 111 encodes the component variable having the dimension number m using the correlation learning engine 102 corresponding to the model ID.
The tag creating unit 112 describes the model ID and the component variable having the dimension number m based on the model ID in a tag.

As a decoding side, the tag reading unit 121 reads a model ID and a component variable having a dimension number m based on the model ID from the tag.
The decoding unit 122 decodes the read component variable having the dimension number m into a three-dimensional model using the correlation learning engine 102 corresponding to the model ID.

<Estimation of missing values of measuring values>
FIG. 13 is an explanatory diagram showing a measurement value space for estimating missing values in the present invention.
FIG. 14 is a simple code representing missing value estimation in the present invention.

The measurement value space represents measurement values of a fixed number of dimensions n (= 10) associated with the three-dimensional model of the teacher data group.
Here, according to the present invention, only k (<n) measurement values are determined for the measurement value of the dimension number n of one body as target data, and the other n−k measurement values are missing. May be. That is, according to the present invention, even if a correlation learning model is constructed from, for example, a 10-dimensional measuring value space by a teacher data group, for example, by inputting only the measuring values of k = 3 measuring points, the number of dimensions m Can be estimated.

According to FIG. 13, the isosurface of the super ellipsoid is represented on the measurement value (10 dimensions) space. A super ellipsoid is a figure obtained by expanding an ellipse to a dimension number n (= 10). An isosurface is a contour map drawn on the dimension (= 10). Here, it is assumed that the variance-covariance matrix C = D ^T * D is known.
A quadratic form representing a hyperellipsoid (x: column vector)
f (x) = x ^T * C ⁻¹ * x
C ^-1: symmetric matrix
* Actually, the variance-covariance matrix is C = D ^T * D / number of teacher data groups
* ^-1 indicates inverse matrix

  According to the present invention, a component variable having the number of dimensions m including other optimized nk measuring values is calculated using k measuring values as a constraint. For this, the Lagrange method of Lagrange multiplier is used.

  Lagrange's undetermined multiplier method is an analysis method that is optimized under a constraint condition. Under the constraint condition that the values of some functions are fixed for some variables, Consider the problem of finding the extrema of a function. For each binding condition, a constant (undefined multiplier, Lagrange multiplier) is prepared, and a linear combination using these as coefficients is considered as a new function (an undefined multiplier is also a new variable). Solve as a value problem.

The point where the function f (x1,..., X _n ) takes an extreme value under the constraint g _j (x ₁ ,..., X _n ) = 0 (j = 1,..., K) about,
F (x ₁ , ..., x _n , λ1, ..., λ _k )
= F (x ₁ , ..., x _n ) + Σλ _j g _j (x ₁ , ..., x _n )
Thus, the following expression is satisfied.
dF / dx _i = 0 (i = 1, ..., n)
dF / dλ _j = 0 (j = 1, ..., k)

In the case of estimating missing values of measurement values, if k measurement values are given, the constraint condition g _j (x) = 0 (j = 1,..., K) It is a linear equation representing a plane and is represented by the following equation.
Linear equation representing an affine hyperplane g _j (x) = n _j ^T * (x−p _j ) = 0
n: Hyperplane normal vector
p: A point on the hyperplane In particular, each hyperplane is orthogonal to the base (the direction of n coincides with the base direction).
g _j (x) = x _i −y _j = 0
y _j : j-th measuring value
x _i : Corresponding element of x Finding the minimum value of the function f (x) under the constraint condition finds the body shape closest to the average under the given measurement value.

Specifically, in the case of a 10-dimensional space, it is expressed as follows.
x: 10-dimensional column vector y: 10-dimensional column vector containing k measurement values (values of measurement values other than k are arbitrary)
λ: k-dimensional column vector whose elements are Lagrange multipliers O: matrix of k rows and 10 columns Each row is a one-hot row vector corresponding to a given measurement value C: variance covariance matrix f (x) = 1/2 * x ^T * C ^-1 * x
g (x) = O * (y−x)
F (x) = f (x) + λ ^T * g (x)
dF / dx = C ⁻¹ * x−O ^T * λ = 0 (1)
dF / dλ = O * (y−x) = 0 (2)
From (1), x = C * O ^T * λ (3)
Substitute (3) for (2)
O * y-O * C * O ^T * λ = 0
λ = (O * C * O ^T ) ⁻¹ * O * y
Substituting λ into (3)
^{x = C * O T * (} O * C * O T) -1 * O * y

FIG. 15 is a user interface representing a three-dimensional model, measurement values, and QR code in the present invention.
According to this user interface, the three-dimensional model corresponding to the measured value and the QR code in which the component variable of the three-dimensional model is described can be seen at a glance. In particular, it is possible to share a three-dimensional model simply by having a QR code read by a camera.

As described above in detail, according to the program and apparatus of the present invention, it is possible to easily generate a three-dimensional model from only measured values and to encode and decode it to an extremely small data capacity.
The present invention is suitable for the following uses, for example.
(Use 1) Using a user-owned smartphone, input a measurement value of his / her body shape and create a QR code including component variables. By holding the QR code over the camera of the transfer destination device, a three-dimensional model representing its body shape can be instantaneously transmitted.
(Use 2) As a use in the clothing industry, a three-dimensional model suitable for the clothes can be created by inputting the measurement value of the clothes.
(Use 3) As a use in the medical industry, by encoding a three-dimensional model based on a user's DICOM data into component variables, a medical three-dimensional model of the user's various organs can be stored in a tag or RFID. Can be transferred instantly.
(Use 4) As a use in the advertising industry, a three-dimensional model can be reproduced from a component variable read by a user, and the three-dimensional model can be displayed on a signage display as the user's character.
(Use 5) The encoded component variable cannot be easily recognized by a third party and has confidentiality of personal information.

  Various changes, modifications, and omissions of the above-described various embodiments of the present invention can be easily made by those skilled in the art. The above description is merely an example, and is not intended to be restrictive. The invention is limited only as defined in the following claims and the equivalents thereto.

1 Device 101 Statistical Learning Engine 102 Correlation Learning Engine 111 Encoding Unit 112 Tag Creation Unit 121 Tag Reading Unit 122 Decoding Unit

Claims (17)

A program for causing a computer mounted on an apparatus for generating a three-dimensional model composed of a plurality of vertices to function from a measuring value of one dimension n as a target data,
As the teacher data group, for each three-dimensional model, dimension values of n dimensions corresponding to a plurality of measurement locations are associated,
A statistical learning engine that outputs a component variable of dimension number m from a plurality of three-dimensional models of a teacher data group and constructs a statistical learning model;
A correlation learning engine that constructs a correlation learning model between a dimension n measurement value and a dimension m component variable from a plurality of three-dimensional models of a teacher data group;
Encoding means for encoding from a measurement value of one dimension number n as target data to a component variable of dimension number m using a correlation learning engine;
A program that causes a computer to function as a decoding unit that decodes a component variable having the number of dimensions m into a three-dimensional model using a statistical learning engine. A program for operating a computer mounted on a device for encoding a three-dimensional model composed of a plurality of vertices,
As the teacher data group, for each three-dimensional model, dimension values of n dimensions corresponding to a plurality of measurement locations are associated,
A statistical learning engine that outputs a component variable of dimension number m from a plurality of three-dimensional models of a teacher data group and constructs a statistical learning model;
A correlation learning engine that constructs a correlation learning model between a dimension n measurement value and a dimension m component variable from a plurality of three-dimensional models of a teacher data group;
A program which causes a computer to function as an encoding means for encoding from a measurement value of one dimension number n as a target data into a component variable of dimension number m using a correlation learning engine. In order to extract the measurement value of dimension number n corresponding to a plurality of measurement locations from the 3D model itself of the teacher data group,
Using a statistical learning engine, a component variable having a dimension number m is output by inputting a three-dimensional model of the teacher data group, and reversely reproduced by inputting the component variable having the dimension number m output in reverse reproduction. 3. The computer according to claim 2, wherein the computer functions so as to output a generated three-dimensional model and to extract a measuring value of the number of dimensions n corresponding to a plurality of measuring locations from the reverse reproduction three-dimensional model itself. program. The encoding means includes
As the target data, only k (<n) measurement values are determined for the measurement value of the dimension number n of one body, and the component variable of the dimension number m is determined even if the other n−k measurement values are missing. To estimate
The computer is caused to function so as to calculate a component variable having a dimension number m including other nk measurement values optimized with k measurement values as a constraint. Program. 5. The program according to claim 4, wherein the encoding means causes the computer to function using a Lagrange method of Lagrange multiplier. 6. The program according to claim 2, further causing a computer to function as tag creation means for creating a tag describing a component variable having an encoded number of dimensions m. 6. A teacher data group is prepared for each of the three-dimensional models of different model types,
A plurality of statistical learning engines and correlation learning engines are provided with different teacher data groups for each model type,
The encoding means uses the correlation learning engine that corresponds to a model based identifiers measurement value of the number of dimensions n, encodes the measuring value of the number of dimensions n to component variable dimensionality m,
The program according to claim 6, wherein the tag creating unit causes the computer to function so as to associate a model type identifier and a component variable having a dimension number m based on the model type identifier in association with each other. . A program for operating a computer mounted on a device for decoding a three-dimensional model composed of a plurality of vertices,
As the teacher data group, for each three-dimensional model, dimension values of n dimensions corresponding to a plurality of measurement locations are associated,
A statistical learning engine that outputs a component variable of dimension number m from a plurality of three-dimensional models of a teacher data group and constructs a statistical learning model;
A correlation learning engine that constructs a correlation learning model between a dimension n measurement value and a dimension m component variable from a plurality of three-dimensional models of a teacher data group;
A program that causes a computer to function as a decoding unit that decodes a component variable having a dimension number m as target data into a three-dimensional model using a statistical learning engine. 9. The decoding unit according to claim 8, wherein the decoding unit causes the computer to derive a dimension value of dimension number n from a component variable of dimension number m as target data using a correlation learning engine. program. 10. The program according to claim 8, wherein the computer is caused to function as tag reading means for reading a component variable having a dimension number m from a tag in which a component variable having a dimension number m is described. A teacher data group is prepared for each of the three-dimensional models of different model types,
A plurality of statistical learning engines and correlation learning engines are provided with different teacher data for each model type,
The tag reading means reads a model type identifier and a component variable having a dimension number m based on the model type identifier from the tag,
11. The decoding means causes a computer to function by decoding a component variable having a dimension number m into a three-dimensional model using a correlation learning engine corresponding to the read model type identifier. Program. 12. The program according to claim 6, 7, 10 or 11, wherein the tag causes the computer to function as a QR (Quick Response) code or an RFID (Radio Frequency IDentifier). The three-dimensional model is shape data of a human body or an object, and is expressed with the same number of vertices with respect to the same object.
The program according to any one of claims 1 to 12, wherein the computer is caused to function so that the number of plural teacher data groups is less than the number of vertices of the three-dimensional model. The statistical learning engine is based on Principal Component Analysis,
The program according to any one of claims 1 to 13, wherein the correlation learning engine causes the computer to function based on a least square method. An apparatus for generating a three-dimensional model composed of a plurality of vertices from a measurement value of the number n of dimensions as a target data,
As the teacher data group, for each three-dimensional model, dimension values of n dimensions corresponding to a plurality of measurement locations are associated,
A statistical learning engine that outputs a component variable of dimension number m from a plurality of three-dimensional models of a teacher data group and constructs a statistical learning model;
A correlation learning engine that constructs a correlation learning model between a dimension n measurement value and a dimension m component variable from a plurality of three-dimensional models of a teacher data group;
Encoding means for encoding from a measurement value of one dimension number n as target data to a component variable of dimension number m using a correlation learning engine;
An apparatus comprising: a decoding means for decoding a component variable having the number of dimensions m into a three-dimensional model using a statistical learning engine. A device for encoding a three-dimensional model composed of a plurality of vertices,
As the teacher data group, for each three-dimensional model, dimension values of n dimensions corresponding to a plurality of measurement locations are associated,
A statistical learning engine that outputs a component variable of dimension number m from a plurality of three-dimensional models of a teacher data group and constructs a statistical learning model;
A correlation learning engine that constructs a correlation learning model between a dimension n measurement value and a dimension m component variable from a plurality of three-dimensional models of a teacher data group;
An apparatus comprising: an encoding unit that inputs a measurement value of one dimension number n as target data and encodes it into a component variable of dimension number m using a correlation learning engine. A device for decoding a three-dimensional model composed of a plurality of vertices,
As the teacher data group, for each three-dimensional model, dimension values of n dimensions corresponding to a plurality of measurement locations are associated,
A statistical learning engine that outputs a component variable of dimension number m from a plurality of three-dimensional models of a teacher data group and constructs a statistical learning model;
A correlation learning engine that constructs a correlation learning model between a dimension n measurement value and a dimension m component variable from a plurality of three-dimensional models of a teacher data group;
An apparatus comprising: a decoding unit that decodes a component variable having a dimension number m as target data into a three-dimensional model using a statistical learning engine.

Images