KR20200068777A - Method for preventing the production of weapons based on three-dimensional printers - Google Patents

Abstract

The present invention relates to a method for preventing 3D printer-based manufacturing of weapons. More specifically, the method comprises the steps of: once a 3D printing model is inputted, extracting geometric features from the 3D printing model; classifying the type of the 3D printing model through predetermined data information on the basis of the extracted geometric features and an improved convolutional neural network (CNN); confirming whether the classified type is included in weaponry; and performing the production if the classified type is not included in weaponry as a result of confirmation, and if the classified type is included in weaponry, not performing the production.

Description

METHOD FOR PREVENTING THE PRODUCTION OF WEAPONS BASED ON THREE-DIMENSIONAL PRINTERS}

The present invention relates to a 3D printer-based weapon production prevention method, and more specifically, to detect a 3D weapon model for 3D (Three-Dimensional) printing to prevent weapon production through a 3D printer. It relates to a method for preventing the production of weapons.

With the development of 3D printers, weapons can be easily printed without the limitations of production managers.

The revolution of 3D printing can help users realize their ideas through digital models using 3D printers. Three-dimensional printing is used in a variety of living areas, including jewelry, footwear, industrial design, architecture, engineering and construction, automotive, aerospace, medical and healthcare industries, education and consumer products. With the development of 3D printing technology, people can search 3D weapon models such as firearms, firearms and knives and print or share unlimited 3D objects with a home 3D printer. Especially with new materials, users can print dangerous weapons and use them to damage them. This leads to concerns about 3D printer security because anyone can print dangerous weapons.

So far, the dangers of weapons manufactured with 3D printers have been proven, but concerns about the dangers of 3D printer weapons have just begun to form into considerations and policies. Researchers are not interested in the aspect of how to limit the printing of 3D weapon models, and there is no solution in the 3D printing industry to stop printing 3D weapon models. Pistol detection technology based on image processing methods applied to surveillance systems or security systems in special places such as airports and buildings cannot be applied to secure 3D printers. The input of the 3D printer is not an image because it is a 3D printing model.

In addition, 3D model matching technology cannot be applied to secure 3D printing to prevent printing of 3D weapon models. This is because the 3D model matching technology must access the database of the 3D model and make a decision. If the model-type sample model is not stored in the database of the 3D model, the input model does not match any model when querying with the 3D model matching technique.

Therefore, for a safe 3D printing industry, a technology capable of detecting a 3D weapon model for 3D printing needs to be developed.

Therefore, the present invention has been proposed to solve the above-described problems, and by detecting a 3D weapon model for 3D printing to prevent weapons from being produced through a 3D printer, it is possible to implement a safe 3D printing industry. It is to provide a method for preventing production of a weapon based on a 3D printer.

In addition, the present invention uses a D2 shape distribution and a deep learning-based convolutional neural network (CNN), a 3D printer-based weapon capable of detecting whether a 3D printing model input to a 3D printer is a 3D weapon model with high accuracy. It is to provide a method for preventing production.

The objects of the present invention are not limited to those mentioned above, and other objects not mentioned will be clearly understood by those skilled in the art from the following description.

A 3D printer-based weapon production prevention method according to the present invention for achieving the above object, when a 3D printing model is input, extracting geometric characteristics from the 3D printing model; Classifying the type of the 3D printing model through pre-trained data information based on the extracted geometric characteristics and an improved convolutional neural network (CNN); Determining whether the classified type is included in a weapon; And, as a result of the verification, when the classified type is not included in the inorganics, production is performed, and when the classified type is included in the inorganics, production is not performed.

According to the present invention, by detecting a 3D weapon model for 3D printing to prevent the production of weapons through a 3D printer, it is possible to implement a safe 3D printing industry.

In addition, according to the present invention, by using a D2 shape distribution and deep learning-based CNN, it is possible to detect whether a 3D printing model input to a 3D printer is a 3D weapon model with high accuracy.

The effects of the present invention are not limited to those mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the following description.

1 is a diagram showing the structure of a 3D triangular mesh of a specific weapon,
2 is a flowchart illustrating a method for performing enhanced CNN-based learning according to an embodiment of the present invention,
3 is a diagram showing an example of a process of calculating a D2 shape distribution from a 3D triangular mesh according to an embodiment of the present invention;
4 is a diagram showing a D2 vector constructed from D2 shape distribution according to an embodiment of the present invention;
5 is a diagram illustrating an example of a process of learning a D2 vector based on an improved CNN according to an embodiment of the present invention,
6A to 6C are histograms of D2 shape distribution for each type of 3D model according to an embodiment of the present invention;
7 is a flowchart illustrating a method for preventing production of a 3D printer-based weapon according to an embodiment of the present invention.

The objectives and effects of the present invention, and technical configurations for achieving them, will be clarified with reference to embodiments described below in detail together with the accompanying drawings. In describing the present invention, when it is determined that a detailed description of known functions or configurations may unnecessarily obscure the subject matter of the present invention, the detailed description will be omitted. In addition, terms to be described later are terms defined in consideration of donations in the present invention, which may vary according to a user's or operator's intention or practice.

However, the present invention is not limited to the embodiments disclosed below, but may be implemented in various different forms. Only the present examples are provided to make the disclosure of the present invention complete, and to fully inform the person of ordinary skill in the art to which the present invention pertains, the scope of the invention being defined by the scope of the claims. It just works. Therefore, the definition should be made based on the contents throughout this specification.

On the other hand, in the specification, when a part is "connected" with another part, it is not only "directly connected" but also "indirectly connected" with another member in between. Includes. Also, when a part “includes” a certain component, this means that other components may be further provided instead of excluding other components, unless otherwise stated.

Conventionally, a 3D model similar to a 3D model input by a user was searched among 3D models stored in a database through 3D model matching technology. However, this 3D model matching technique has a disadvantage in that when a 3D model input by a user is not stored in a database, there is a fear of searching for an incorrect 3D model, and its accuracy is also low.

Therefore, it is not possible to detect whether the 3D printing model input to the 3D printer is a 3D weapon model through the 3D model matching technology.

The user needs to input a 3D printing model into a 3D printer in order to produce a 3D stereoscopic article through a 3D printer, where the 3D printing model is a 3D triangular mesh designed by CAD software.

1 is a view showing the structure of a 3D triangular mesh of a specific weapon.

Referring to FIG. 1, the 3D triangular mesh includes a facet set. Each facet contains three vertices and a normal vector. Each vertex is represented by three coordinates: x, y and z.

The present invention recognizes the D2 shape distribution obtained from the surface of the 3D triangular mesh, and learns the 3D weapon model with high accuracy by learning the D2 vector composed of the D2 shape distribution based on the improved CNN. Thus, it is possible to prevent the production of a weapon through a 3D printer.

Hereinafter, a 3D printer-based weapon production prevention method according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

2 is a flowchart illustrating a method of performing improved CNN-based learning according to an embodiment of the present invention.

First, 3D printing models for weapons and non-weapons, that is, 3D triangular meshes are collected, and improved CNN-based learning is performed based on the collected 3D triangular meshes.

Referring to FIG. 2, when a 3D triangular mesh is input, N pairs of two arbitrary points are generated by extracting facets and vertices from the input 3D triangular mesh (S201). Here, the two arbitrary point pairs are randomly selected from points on the surface of the 3D triangular mesh including the vertices of the 3D triangular mesh.

Thereafter, a distance is calculated from two arbitrary points of N pairs generated in step S201 (S203), and a D2 shape distribution of the 3D triangular mesh is calculated through the calculated distance (S205). After calculating the D2 shape distribution in step S205, a D2 vector is constructed from the histogram of the D2 shape distribution (S207).

Meanwhile, the type is classified through the D2 vector (S209), and the weight is updated according to the similarity degree of the D2 vector in the classified same type (S211). Through this, the learning result for the D2 vector can be improved.

Hereinafter, each step described in FIG. 2 will be described in detail with reference to FIGS. 3 to 5.

3 is a diagram illustrating an example of a process of calculating a D2 shape distribution from a 3D triangular mesh according to an embodiment of the present invention.

The D2 shape distribution of a 3D triangle mesh is a distribution of a set of distances calculated from any two sets of points on the surface of the 3D triangle mesh. To calculate the D2 shape distribution from a 3D triangle mesh, first, two surfaces of the 3D triangle mesh are After generating N pairs of random points, calculate the distance from the N pairs.

As described above, the 3D triangular mesh includes a set of triangles, each of which has three vertices. That is, consider 3D triangular mesh M = {V, F}. Here, V is a set of vertices, F is a set of facets, and can be expressed as Equation 1 below.

<Equation 1>

는

로부터 형성된다.here,

The

Is formed from.

Meanwhile, N pairs of two arbitrary points are generated from a set of vertices and a set of facets, and can be expressed as Equation 2 below.

<Equation 2>

는 {V, F}에서 임의로 선택된 두 점이다.here,

Are two randomly selected points in {V, F}.

을 계산하며, 여기서,

는

사이의 유클리드 거리이며, 하기 <수학식 3>과 같이 나타낼 수 있다.On the other hand, a set of distances from two random points of N pairs

Calculate, where:

The

It is the Euclidean distance between and can be expressed as <Equation 3> below.

<Equation 3>

및

라 가정하면, 하기 <수학식 4>와 같이 나타낼 수 있다.Next, find the minimum and maximum distances in this set of distances to determine the range of values for the distance. This minimum distance and maximum distance respectively

And

Assuming D, it can be expressed as Equation 4 below.

<Equation 4>

,

)를 B빈으로 나눔으로써 3D 삼각형 메쉬의 D2 형상 분포를 계산하고, 각 빈에 속하는 거리의 수를 센다. 각각의 빈의 평균 폭 및 각 빈의 값 범위는 하기 <수학식 5>와 같이 나타낼 수 있다.<수학식 5>after, (

,

) Is divided by B bin to calculate the D2 shape distribution of the 3D triangular mesh and count the number of distances belonging to each bin. The average width of each bin and the value range of each bin can be expressed as Equation 5 below.

는 상응하는

거리를 갖는다고 가정한다. 따라서, 3D 삼각형 메쉬의 D2 형상 분포는 하기 <수학식 6>과 같이 B빈들로 표현되고, 각 빈

는

거리를 갖는다.Each blank

Is equivalent

Assume you have a distance. Therefore, the D2 shape distribution of the 3D triangular mesh is represented by B bins as shown in <Equation 6> below, and each bin

The

Have a distance

<Equation 6>

를 각 빈

의 대표 값으로 간주하면 상기 <수학식 6>은 하기 <수학식 7>과 같이 된다.

Each bin

Regarding as a representative value of <Equation 6> is as follows <Equation 7>.

<Equation 7>

4 is a diagram showing a D2 vector constructed from a D2 shape distribution according to an embodiment of the present invention.

As described above, after calculating the D2 shape distribution, the D2 vector is constructed from the histogram of the D2 shape distribution.

The number of bins is the number of elements in the corresponding D2 vector. Accordingly, the D2 vector is a set of B elements, and is calculated by Equation 8 below.

<Equation 8>

는 빈

에 있는 거리의 수이다.도 5는 본 발명의 실시예에 따라 D2 벡터를 향상된 CNN 기반으로 학습하는 과정의 일 예를 나타내는 도면이다.here,

Is empty

5 is a diagram illustrating an example of a process of learning a D2 vector based on an improved CNN according to an embodiment of the present invention.

The D2 vector is trained based on the enhanced CNN, where the structure of the enhanced CNN consists of one convolution layer and a neural network, as shown in FIG. 5.

The input of the convolution layer is a D2 vector, each D2 vector is a set of B discrete elements, and the output of the convolution layer is an input neuron for a neural network.

은 컨볼루셔널 레이어의 컨볼루션 함수이고,

은 컨볼루션 프로세스의 출력이라고 가정하면, 입력 뉴런

은 하기 <수학식 9>와 같이 나타낼 수 있다.

Is the convolution function of the convolutional layer,

Assuming is the output of the convolution process, the input neuron

Can be expressed as <Equation 9> below.

<Equation 9>

Here, g is the kernel of convolution, and m is the length of g.

을 학습한다. 역 전파 알고리즘은 순방향 통과, 손실 함수, 역방향 통과 및 가중치 갱신의 네 가지 섹션으로 구분된다. 순방향 통과 동안

은 전체 네트워크를 통과한다. 하이퍼볼릭 탄젠트 함수는 히든 레이어(Hidden Layer)에 활성 함수로 적용되고, 소프트-맥스(soft-max) 함수는 출력 레이어(Output Layer)에 활성 함수로 적용된다. 손실 함수는 신경망의 출력 오류를 계산하는데 사용된다. 목표 출력과 실제 출력에서 계산된다. 손실 함수는 다양한 방법으로 정의할 수 있지만 MSE(Mean Squared Error)가 일반적이다. 신경망은 MSE 함수에 의해 계산된 오차에 기초하여 역방향 통과 및 가중치 갱신 프로세스를 통해 히든 레이어의 가중치 및 바이어스를 조정하고 갱신한다.In the course of training by neural networks, input neurons are used using the inverse propagation algorithm.

To learn. The back propagation algorithm is divided into four sections: forward pass, loss function, reverse pass and weight update. During forward passage

Goes through the entire network. The hyperbolic tangent function is applied to the hidden layer as an active function, and the soft-max function is applied to the output layer as an active function. The loss function is used to calculate the output error of the neural network. It is calculated from the target output and actual output. The loss function can be defined in various ways, but Mean Squared Error (MSE) is common. The neural network adjusts and updates the weight and bias of the hidden layer through a reverse pass and weight update process based on the error calculated by the MSE function.

6A to 6C are histograms of D2 shape distribution for each type of 3D model according to an embodiment of the present invention, and FIGS. 6A and 6B are all for weapons, respectively, for 3D models of firearms and knives. Is a histogram of the D2 shape distribution for, and FIG. 6C is a histogram of the D2 shape distribution for a 3D model of a three-dimensional article having a plane, church, and Batman shape as for non-weapon.

6A to 6C, it can be seen that even if 3D models of the same type, the number of facets, the number of faces, and the like in the 3D triangular mesh are different from each other. However, it can be confirmed that all D2 shape distributions of the same type of 3D model are similar, which means that the D2 vectors of each 3D triangular mesh of the same type are similar. In addition, it can be confirmed that the D2 shape distributions of different types, that is, the inorganic model and the non-inorganic model are different from each other, which means that the D2 vectors of the inorganic and non-inorganic models are different. On the other hand, if the D2 vectors in the same type are similar, the result of learning is improved.

That is, by learning and using a D2 vector composed of a D2 shape distribution based on an improved CNN, it is possible to detect whether a 3D printing model input to a 3D printer is a weapon or a non-weapon.

7 is a flowchart illustrating a method for preventing production of a 3D printer-based weapon according to an embodiment of the present invention.

When a 3D printing model, that is, a 3D triangle mesh, is input to the 3D printer (S701), N pairs of two arbitrary points are generated by extracting facets and vertices from the input 3D triangle mesh (S703). Here, the two arbitrary point pairs are randomly selected from points on the surface of the 3D triangular mesh including the vertices of the 3D triangular mesh.

Thereafter, a distance is calculated from two arbitrary points of N pairs generated in step S702 (S705), and a D2 shape distribution of the 3D triangular mesh is calculated through the calculated distance (S707). After calculating the D2 shape distribution in step S705, a D2 vector is constructed from the histogram of the D2 shape distribution (S709).

Based on the D2 vector and the improved CNN, the type is classified through pre-trained data (S711). That is, it is to classify what type of model the 3D printing model input in step S701.

Thereafter, it is determined whether the 3D printing model is a weapon through the classification result (S713).

On the other hand, as a result of the determination in step S713, when it is determined that the 3D printing model is non-inorganic, production is performed (S715), and when it is determined that the 3D printing model is weapon, non-production is performed (S717).

Accordingly, the present invention is to determine whether the 3D printing model is a weapon based on the geometric characteristics of the 3D printing model, thereby preventing the production of weapons through the 3D printer.

In the present specification and drawings, preferred embodiments of the present invention have been disclosed, and although specific terms are used, they are merely used in a general sense to easily describe the technical contents of the present invention and to understand the present invention. It is not intended to limit the scope. It is apparent to those skilled in the art to which the present invention pertains that other modifications based on the technical spirit of the present invention can be implemented in addition to the embodiments disclosed herein.

Claims (4)

In the 3D printer-based weapon production prevention method,
When a 3D printing model is input, extracting geometric characteristics from the 3D printing model;
Classifying the type of the 3D printing model through pre-trained data information based on the extracted geometric characteristics and an improved convolutional neural network (CNN);
Determining whether the classified type is included in a weapon; And
As a result of the check, when the classified type is not included in the weaponry, the production is performed, and when the classified type is included in the weaponry, the production is not performed. Way.
According to claim 1,
Extracting the geometric properties from the 3D printing model,
Generating two arbitrary points of N pairs in the 3D printing model;
Calculating a distance from the two arbitrary points of the N pair;
Calculating a D2 shape distribution through the calculated distance; And
And constructing a D2 vector from the histogram of the D2 shape distribution.
According to claim 2,
The two arbitrary points of the N pair,
A method for preventing production of a weapon based on a 3D printer, characterized in that it is randomly selected from points on a surface of a 3D printing model including the vertices of the 3D printing model.
According to claim 2,
The D2 shape distribution,
A 3D printer-based weapon production prevention method, characterized in that it is a distribution of a set of distances calculated from a set of arbitrary two points on the surface of the 3D printing model.

Images