WO2020217082A1 - Generate editable 3d object files on the basis of two-dimensional images based on deep learning n gan network (generative adversarial networks) - Google Patents

Abstract

Generate editable 3D object files based on multiple two-dimensional images photographed from different views of an object using Machine Vision, Blender API and Machine Learning based on GAN neural networks for texture making and final view generation in short times. Object different views images, captured in a studio that contains a frame, three frame fixed square plates for holding lights and camera in three directions and a rotating plate in the center. Captured images processed in designed software. Object boundary detected using SSD algorithm. Generate four RGB-D matrices from different four light source on each direction plate, combined created matrices and calculate object faces normal-border in three dimensional, finally using blender API, calculated object pixel profiled matrix convert to the raw 3D object model in Wavefront format and textured with our GAN learned network result.

Description

Description

Title of Invention : Generate editable 3D object files on the basis of two-dimensional images based on deep learning in GAN network (GENERATIVE ADVERSARIAL

NETWORKS)

[0001 ] Generate editable 3D object files based on multiple two-dimensional images photographed from different views of an object using Machine Vision, Blender API and Machine Learning based on GAN neural networks for texture making and final view generation in short times. Object different views images, captured in a studio that contains a frame, three frame fixed square plates for holding lights and camera in three directions and a rotating plate in the center.

[0002] Captured images processed in designed software. Object boundary detected using SSD algorithm. Generate four RGB-D matrices from different four light source on each direction plate, combined created matrices and calculate object faces normal-border in three dimensional, finally using blender API, calculated object pixel profiled matrix convert to the raw 3D object model in Wavefront format and textured with our GAN learned network result.

Technical Field

[0003] The present invention relates generally to COMPUTING; CALCULATING;

COUNTING - MAGE DATA PROCESSING OR GENERATION and Artificial intelligence .

Background Art

[0004] Description or the related art including information disclosed . examples of such assemblies are disclosed in the following U.S. pat. Nos.:

[0005] Manual Training Assist Modeler for Three-dimensional Volumes - Registration Number: 83137 :

[0006] With the Universal Modeler, diverse industrial and educational 3D models at the shortest time and at the lowest cost can be made. The device has 4 axes, with 2 linear axes and two other rotational axes, and cellular cutting, it is done with heating the element. The existence of linear and rotational axes allows different three-dimensional parts to be built and used in teaching. The outstanding feature of this design, which distinguishes it from other cellular cutters, is the use of wire elements that can be formed. Application of formed elements provides the possibility of making spherical bodies, creating impasse and seduced holes in various forms, building cones and cylinders by creating a hole in its axis in impasse and seduced form, making a ring cylinder, creating sharp and arch corners and edges. Each device is made based on some features. For example, the basis of the cutting machine is the manufacturing of cylindrical parts. With the device, it is possible to build any three-dimensional part whose foundation is cylindrical, conical, spherical, pyramidal, charter and cubic and it can be used as a teaching aid in the classroom for better and faster explanation and visualization of technical drawing lessons whose teaching is not possible without the presence of real pieces. The main applications of the device include the construction of three-dimensional models, the making of casting models, the creation of plates and seventh lines of technical drawing, the development of hollow cylindrical and conical or pyramidal parts, the construction of the building replica, and the best tool for teaching conic sections and interface of objects.

[0007] The stated design is the construction of three-dimensional model and in

general 3D printing, while our proposed design is a system based on artificial intelligence and the LEARNING machine and BLENDER software that the editable three-dimensional files are extracted from two-dimensional images.

[0008] Providing of a set of techniques for the creation and adaptation of three- dimensional software models and mechanical tests of femoral bones under physiological loads - Registration Number: 68905 :

[0009] Due to the diverse use of the three-dimensional model of femur bone in

medicine and biomedical engineering, as well as the necessity of adapting geometry and loading this model with the actual sample, a set of methods and equipment is recommended as follows. It includes the definition of a plate and a base axis in the femur bone which are transmitted to software models and fixture of mechanical tests by a frame and a disc attached to the bone. In practice, the three-dimensional model of the femoral bone is created by performing image processing techniques on images obtained from a bone collection, disc, and reference plate CT scan. Thus, by placing a bone and a disk in a three- dimensional fixture of a mechanical test, the possibility of applying physiological loads in different spatial directions and adapting this loading with software loading of the three-dimensional model is provided.

[0010] The mentioned design is a three-dimensional design for the femur bone used for mechanical testing, while our proposed design is to extract editable three- dimensional file, which is used and edited in three-dimensional software, if necessary.

[0011 ] EMPLOYING THREE-DIMENSIONAL (3D) DATA PREDICTED FROM TWO- DIMENSIONAL (2D) IMAGES USING NEURAL NETWORKS FOR 3D

MODELING APPLICATIONS AND OTHER APPLICATIONS - United States Patent Application 20190035165A1 :

[0012] The obvious subject focuses on the application of machine learning models adapted to predict the three-dimensional data from two-dimensional images using deep learning techniques to obtain three-dimensional data for two-dimensional images. In some reforms, a method is proposed that includes accepting an operating system coupled to the processor, a two-dimensional image, and determining the auxiliary data for a two-dimensional image by the system, where auxiliary data contains orientated information about the recorded image orientation of the two-dimensional image. Moreover, the method includes the result obtained by the system using one or more neural network models configured to obtain 3D information based on a two-dimensional image and auxiliary data.

[0013] The design merely transforms the two-dimensional images into a three- dimensional images, but converting two-dimensional images into three- dimensional editable files is performed in our proposed design.

[0014] Image segmentation using neural network method - United States Patent 9947102 :

[0015] The present disclosure relates to systems, methods, devices, and stable

computer readable medium storage for the segmentation of three-dimensional images. In one implementation, a computer-implemented method is proposed for the segmentation of a three-dimensional image. The method may include receiving a three-dimensional image obtained by an imaging device and selecting a number of two-dimensional images adjacent to a three-dimensional image.

[0016] The design is to separate 3D images for two-dimensional images, while our proposed design is to extract editable 3D files from two-dimensional imaging.

[0017] Evaluation of three-dimensional scenes using two-dimensional

representations - United States Patent 9111375 :

[0018] A compatible system for implementing the learning law in a three-dimensional (3D) environment is described. The system includes: Compatible rendering to produce a two-dimensional (2D) image that is at least partly in a 3D scene; a computational element adapted to produce a series of physical properties based on minimum two-dimensional images; and a classification of adapted properties to produce at least one set of learned attributes based on at least part of the set of physical properties and to produce estimated scene features based on the least part of the set of features learned. A Label method is any image from a set of two-dimensional images with scene information directed to a three-dimensional scene.

[0019] In the design, a rendering is also taken from two-dimensional images and is used in a three-dimensional plate, while the nature of the proposed design is different from all of these designs and our main goal is to create editable three- dimensional files.

[0020] Depth sensing using an RGB camera- United States Patent 9626766 :

[0021 ] It is a depth sensing method using an RGB camera. In a method, for example, the image color of a scene is received from an RGB camera. The color image is part of the training machine learning, so that the features of the image elements are used to refer to all or part of the image elements of the depth value, which represents the distance between the marker surface by the image element and the RGB camera. In different samples, the machine learning component consists of one or more geographically approved patchwork forest.

[0022] The design has been taken from two-dimensional images of depth value by the LEARNING machine, while background of the images is removed in the design presented with the LEARNING machine, the background image of the image is removed and the original model of the three-dimensional file without background is created.

[0023] Evaluation of Three-Dimensional Scenes Using Two-Dimensional

Representations- United States Patent Application 20100014781 :

[0024] For example, a method of converting two-dimensional images into three- dimensional images is a readable medium for the provided computers and systems. These visualizations can be produced based on an image in the database with depth information or any depth information. Given that the two- dimensional image becomes three-dimensional content, the background image adapted from databases is formed. In addition, graphical analysis and

comparisons of techniques for detecting a two-dimensional image preview are used, so that the corresponding depth map can be generated from the

foreground or background information. So three-dimensional content can be created with a two-dimensional image with depth information. As a result, users can quickly gain three-dimensional content automatically from the two- dimensional image and three-dimensional content provided.

[0025] In this design, three-dimensional images are created from two-dimensional images, but they are not editable. While two-dimensional images are

photographed in the proposed design, and then two-dimensional images are transformed into editable three-dimensional file by learning machine and blender software, and, in none of the designs, the issue is referred to.

Summary of Invention

[0026] This summary is intended to provide an overview of the subject matter of the present disclosure, and is not intended to identify essential elements or key elements of the subject matter, nor is it intended to be used to determine the scope of the claimed implementations. The proper scope of the present disclosure may be ascertained from the claims set forth below in view of the detailed description below and the drawings.

[0027] In one general aspect, the present disclosure describes Generate editable 3D object files on the basis of two-dimensional images based on deep learning in GAN network in the firld of PHYSICS (G) - COMPUTING; CALCULATING , COUNTING , MAGE DATA PROCESSING OR GENERATION, IN GENERAL and Artificial intelligence .

[0028] The possibility of providing 3D models of goods for commercial purposes and making computer games is one of the most important needs. Generally, there are 2 methods for making 3D editable models and applicable in modeling programs: The first method is the designing of a three-dimensional model in special software for modeling and using the model made in the required cases, and the second method is the reconstruction of a 3-D model from the existing objects that their three-dimensional computer model is not available. The first method can be used in many cases. In the second method, due to its inherent limitations, it has been less considered for commercial use to date. Now, using 3D advanced scanners, 3D models of objects can be created. But this use of this method has three major problems, including the high cost of using this method, the problem of the dimensions of objects that require large-scale scanners for large dimensions and the displacement of objects or 3D scanners to the locations of objects. One of the main reasons for lack of commercializing these methods is the problem of generating the final 3-D model, which is obtained in the form of an image and cannot be edited in 3D modeling programs. The editable 3D files based on multiple images photographed has two-dimensional depths from different views of an object using Machine Vision, and its automatic reconstruction using the Blender API and texture making and the final view is on the basis of Machine Learning based on GAN neural networks in time less than 30 seconds.

Technical Problem

[0029] Three-dimensional printers can be used in making simple everyday tools and producing human organs or using in space industries. Today, 3D modeling is common in a variety of fields such as parts manufacturing, architecture, industrial design, robotics, aerospace industries, and more. These models were previously presented in the form of two-dimensional images on screen or on paper so that people get an understanding of what designers have in their minds. Three- dimensional printers have the ability to produce any kind of parts with any shape and angle, solid or hollow, straight or curved, for each parts with any design. The need is palpable everywhere. Industry, medicine, education, automotive, military, and everything that needs to simulate, manufacture replica and build a prototype, using a three-dimensional printer, can accelerate the time-consuming process of simulating and manufacturing parts replica and examine the parts only by printing a three-dimensional design in very little time. Ability to prepare 3D models of goods for commercial purposes and making computer games is one of the very important needs. Generally, there are 2 methods for making 3D editable models and applicable in modeling programs: The first method is the designing of a three-dimensional model in special software for modeling and using the model made in the required cases, and the second method is the reconstruction of a 3D model from the existing objects that their three-dimensional computer model is not available. The first method can be used in many cases, but this process will be very time consuming and costly in the face of the large volume of objects that the need for their modeling is required. Although the models made will have a very high accuracy in detail, but generally this amount of detail is required only in highly precise industrial applications, and models with less details can be also used in commercial applications and making computer games. The second method, due to its inherent limitations, has been less considered for commercial use to date and it is used more in the field of research or in very limited cases. The main limitation of this method is the conversion of two-dimensional images into three-dimensional ones. Now, using 3D advanced scanners, 3D models of objects can be created. But this use of the method has three major problems, including the high cost of using this method, the problem of the dimensions of objects that require large-scale scanners for large dimensions and the

displacement of objects or 3D scanners to the locations of objects.

[0030] Based on the existence of these problems, new research has begun to

convert the images of photographic cameras into two-dimensional objects.

Numerous laboratory and research cases have been published for this purpose in recent years. In these methods, using learning machine and deep learning in the field of neural networks and productive networks, after learning, computer attempts to reconstruct the three-dimensional image of many examples. Although the principle of technology is very advanced and building a 3D image of a two- dimensional image is a very important achievement in these cases, it has not yet come to the step of commercial use. The main reasons for lack of

commercializing these methods is the problem of producing the final 3-D model, which is obtained in the form of an image and cannot be edited in 3D modeling programs. Moreover, the final models have a complete dependence on the models trained in the system, and the final image quality cannot be used in most commercial applications. Another common method for producing 3D models of objects by the image is the use of stereo photography technique. Although this method has good quality in comparison with other approaches, but due to the use of only one direction from the view of the objects, the final three-dimensional model at the viewing angles other than front side has no proper structure and good quality. In the proposed method of this design, we have combined different methods for performing each phase from a set of phases to produce three- dimensional objects, and finally, using the Blender software and the learning machine and simultaneous imaging of the target object, it is possible to produce editable three-dimensional files.

Solution to Problem

[0031 ] The hardware infrastructure of the design is as follows:

[0032] For this section, an innovative design for imaging has been used. According to the original method of construction on the basis of identifying the strengths of the depth image-based model, one of the following methods will be imaged using the special raspberry pi platform cameras, along with 4 different light sources in 3 directions, top, right and front of three-dimensional objects. So, using 3 cameras and 4 fixed light sources for each camera in a frame with the ability to change the dimensions, each camera and source of light in each direction will be controlled by the raspberry pi platform. Also, a robotic arm capable of moving in 3 directions, top, front and right, with a camera and 4 light sources controlled by the raspberry pi platform can be used. A red laser light source is used below each camera to adjust the position of the object in the middle coordinates of the set.

[0033] The software infrastructure of the design is as follows:

[0034] In the software section during the following different processes, two- dimensional image data with three views is extracted and finally the editable model in three-dimensional modeling programs will be produced: Identifying the object in the image using the SSD algorithm (Given the possibility of imaging the free space of the environment without the existing objects, if necessary, the background image will be removed with the help of different algorithms).

Calculating a spatial position vector for each pixel of the image in each view based on 4 D-RGB color depth models. Making the matrix of each view based on the position calculations of each pixel and integrating the matrix of 3 views to extract the final position of each pixel with the normal vector. Making the Wave front.obj file and creating textures by images extracted from 3 views using the DCGAN neural network on the final model.

[0035] Hardware details and images

[0036] Camera: Raspberry Pi Camera Module 72, 8 megapixel sony IMX219 image sensor

[0037] Camera controller and light and laser sources: Raspberry Pi Zero W

[0038] Light Source: LED

[0039] Laser Light Source: TTL LASER DIODE RED

[0040] Connection between components: WiFi

[0041 ] Image resolution: pixels 2464 * 3280

[0042] Image format: 10-bit RAW RGB data

[0043] Executive process:

[0044] All parts used in Blender software are conducted using the Python

programming language and do not require human resources.

[0045] The first step in the implementation process is the establishment of imaging cameras or robotic arm. In the case of lacking the use of robotic arm, it is necessary that cameras in each view are deployed at the shooting location. The location of the cameras will be on an aluminum frame with the ability to change the size in telescopic form and the ability to rotate the object by rotational screen that the object will be placed on it - If required, multiple views - (or use of a robotic arm with the ability to move in all directions having a plate). The cameras with 4 light sources and a laser light source pointer are placed on a square- shaped plate. The camera at the center of the plate and 4 light sources in the center of the sides and the laser light source pointers are also at the closest distance of bottom of the camera at the center of the plate.

[0046] The imaging will be done by the device in three directions from the top, from the front, from the right or left of the desired object located in the center of the collection. Due to the importance of deploying the object in the middle of the field of view, the cameras will be more accurately provided in the center of the collection to achieve better quality and accuracy using 3 laser pointers that are projected from the 3 directions on the object. After the exact location of the object in the center of the collection, the computer of controller will connect to any of the raspberry pi controllers via Wi-Fi connection and will perform the initial

configuration required. After the preparation of 3 sets of imaging of 3 directions or a single set deployed on the robotic arm, the system will be ready to image.

[0047] In the first step, imaging from any direction creates a raw image without the presence of object from the environment as a background image. In the next step, 4 images of the object are taken in each direction by the order, depending on the presence of 4 different light sources. To apply the texture on the final three dimensional object in each view, an image will be taken simultaneously using each of the four light sources. Finally, at the end of the imaging process, 18 images of the object in question are created.

[0048] In the next step, the images in the designed servers will enter the GPU

processing stage. In this step, the location of the object in the image will be initially identified using the SSD algorithm that is trained according to the previous data, and then 18 directions in pure form without background will be prepared using the algorithms to remove the background image by combining the existing background image from the pre-existing image of the object in all the images.

The images are purified using different visual machine techniques based on a diagnostic analytical model of depth in the images for each direction, 4 color depth matrices will be calculated. After calculating the color depth matrix, the necessary normal vectors are extracted, and the spatial position of each pixel of the image will be determined in the form of a pixel with dimensions of 1 unit in 1 unit in 1 unit in spatial coordinates. Through the hybrid calculation of 4 image color depth in any direction, these spatial coordinates lead to a final matrix for each direction, and a zero dimensional 3D matrix will be constructed based on the number of pixels in the images made. Then, each point of this matrix is done based on the spatial equations and 3-D data obtained from the 3 past directions of an adaptation process. This matrix is moved to the blender software, and, after a three-dimensional object construction in the blender, a raw wave front file without material is rendered in three different views.

[0049] In the next step, using the DCGAN neural network (previously trained by a large number of examples), we will enter special images of textures in each of the three views, after converting it into a map format, along with a 3D file format to the DCGAN network. Finally, the computational output of the DCGAN network will be a proper file map as the material for the final 3D file, and ultimately, using the blender, the final three-dimensional file and the previous step material will be combined and an editable three-dimensional file will be extracted.

Advantageous Effects of Invention

[0050] 1 -Very low cost in manufacturing

[0051 ] 2-Lack of need for chromatic imaging to remove background

[0052] 3-Lack of need for skilled manpower

[0053] 4-Ability to use for all dimensions of objects

[0054] 5-Possibility of comfortable movement

[0055] 6-High production speed

[0056] 7-Lack of side costs

Brief Description of Drawings

[0057] For a better understanding of the invention and to show how it may be

performed, a preferred embodiment will now be described by way of non-limiting example only, by reference to the accompanying diagrams.

[0058] The drawings show embodiments of the disclosed subject matter for the

purpose of illustrating the invention. However , it should be understood that the present application is not limited to the precise arrangements and

instrumentalities ahown in the drawings , wherein : Fig.1

[0059] [Fig.1 ] is the square-shaped section view that is installed on the stereo and includes a camera and light sources , consistent with one or more exemplary embodiments of the present disclosure.

Fig.2

[0060] [Fig.2] is the Raspberry Pi Camera Module 72 camera that is connected to the stereo frame and includes three cameras in this design , consistent with one or more exemplary embodiments of the present disclosure.

Fig.3

[0061 ] [Fig.3] illustrates each camera contains 4 LED light sources , consistent with one or more exemplary embodiments of the present disclosure.

Fig.4

[0062] [Fig.4] illustrates At the bottom of each camera and the exact alignment of the object, the laser pointer is used, which consists of 3 lasers in this design , consistent with one or more exemplary embodiments of the present disclosure.

Fig.5

[0063] [Fig.5] illustrates a the rear view of the studio which contains aluminum

frames and three plates in three directions , consistent with one or more exemplary embodiments of the present disclosure.

Fig.6

[0064] [Fig.6] illustrates the front view of the studio, consistent with one or more exemplary embodiments of the present disclosure.

DETAILED DESCRIPTION OF THE DRAWINGS

[0065] In the following detailed description, numerous specific details are set forth by way of examples in order to provide a thorough understanding of the relevant teachings. However, it should be apparent that the present teachings may be practiced without such details. In other instances, well known methods, procedures, components, and/or circuitry have been described at a relatively high-level, without detail, in order to avoid unnecessarily obscuring aspects of the present teachings. [0066] The following detailed description is presented to enable a person skilled in the art to make and use the methods and devices disclosed in exemplary embodiments of the present disclosure. For purposes of explanation, specific nomenclature is set forth to provide a thorough understanding of the present disclosure. However, it will be apparent to one skilled in the art that these specific details are not required to practice the disclosed exemplary embodiments.

Descriptions of specific exemplary embodiments are provided only as

representative examples. Various modifications to the exemplary

implementations will be readily apparent to one skilled in the art, and the general principles defined herein may be applied to other implementations and

applications without departing from the scope of the present disclosure. The present disclosure is not intended to be limited to the implementations shown, but is to be accorded the widest possible scope consistent with the principles and features disclosed herein.

[0067] For purposes of reference, it should be understood that the techniques and systems disclosed herein are applicable to coupled motion in a wrist; however, the techniques and systems may be adapted to a number of other applications.

[0068] The device can be used in fashion, so that an editable three-dimensional file of any object can be obtained in less than 30 seconds, and this nature of the file can be edited at any time and there is no need to build and design a sample from the beginning.

[0069] FIG. 1 illustrates a The square-shaped section view that is installed on the stereo and includes a camera and light sources., according to an embodiment herein. With respect to FIG. 1, The square section that the camera and the light source and laser pointer are connected to it 1 , Camera Raspberry Pi Camera Module 72, 8 megapixel sony IMX219 image sensor 2 , 4 LEDs 3 , Laser pointer TTL LASER DIODE RED 4 .

[0070] FIG. 5 illustrates a The rear view of the studio which contains aluminum

frames and three plates in three directions according to an embodiment herein. With respect to FIG. 5, The square section that the camera and the light source and laser pointer are connected to it 1 , Camera Raspberry Pi Camera Module 72, 8 megapixel sony IMX219 image sensor 2 , 4 LEDs 3 , The studio frame that the square plates are connected in the center of it 5 .

[0071 ] FIG. 6 illustrates a The front view of the studio according to an embodiment herein. With respect to FIG. 6 , The square section that the camera and the light source and laser pointer are connected to it 1 , 4 LEDs 3, The studio frame that the square plates are connected in the center of it 5 .

[0072] The separation of various components in the examples described above

should not be understood as requiring such separation in all examples, and it should be understood that the described components and systems can generally be integrated together in a single package, or into multiple systems.

[0073] While the foregoing has described what are considered to be the best mode and/or other examples, it is understood that various modifications may be made therein and that the subject matter disclosed herein may be implemented in various forms and examples, and that the teachings may be applied in numerous applications, only some of which have been described herein. It is intended by the following claims to claim any and all applications, modifications and variations that fall within the true scope of the present teachings.

[0074] Unless otherwise stated, all measurements, values, ratings, positions,

magnitudes, sizes, and other specifications that are set forth in this specification, including in the claims that follow, are approximate, not exact. They are intended to have a reasonable range that is consistent with the functions to which they relate and with what is customary in the art to which they pertain.

[0075] The scope of protection is limited solely by the claims that now follow. That scope is intended and should be interpreted to be as broad as is consistent with the ordinary meaning of the language that is used in the claims when interpreted in light of this specification and the prosecution history that follows and to encompass all structural and functional equivalents. Notwithstanding, none of the claims are intended to embrace subject matter that fails to satisfy the requirement of Sections 101 , 102, or 103 of the Patent Act, nor should they be interpreted in such a way. Any unintended embracement of such subject matter is hereby disclaimed. [0076] Except as stated immediately above, nothing that has been stated or illustrated is intended or should be interpreted to cause a dedication of any component, step, feature, object, benefit, advantage, or equivalent to the public, regardless of whether it is or is not recited in the claims.

[0077] It will be understood that the terms and expressions used herein have the ordinary meaning as is accorded to such terms and expressions with respect to their corresponding respective areas of inquiry and study except where specific meanings have otherwise been set forth herein. Relational terms such as first and second and the like may be used solely to distinguish one entity or action from another without necessarily requiring or implying any actual such

relationship or order between such entities or actions. The terms“comprises,” “comprising,” or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by“a” or“an” does not, without further constraints, preclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0078] The Abstract of the Disclosure is provided to allow the reader to quickly

ascertain the nature of the technical disclosure. It is submitted with the

understanding that it will not be used to interpret or limit the scope or meaning of the claims. In addition, in the foregoing Detailed Description, it can be seen that various features are grouped together in various implementations for the purpose of streamlining the disclosure. This method of disclosure is not to be interpreted as reflecting an intention that the claimed implementations require more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive subject matter lies in less than all features of a single disclosed implementation. Thus the following claims are hereby incorporated into the Detailed Description, with each claim standing on its own as a separately claimed subject matter.

Industrial Applicability [0079] The device can be also used in various industries, including automotive and aerospace, gold and jewelry making, computer games, dentistry, as well as in other industries such as electronics, toy construction, fashion, and, generally, it is used in all industries that need to build prototypes and build sophisticated and editable samples.

Claims

Claims

[Claim 1] generate editable 3D object files based on multiple images photographed with two-dimensional depths from different views of an object using Machine Vision and its automatic reconstruction using the Blender API and texture making and the final view on the basis of Machine Learning based on GAN neural networks in time less than 30 seconds.

[Claim 2] According to claim 1 , a studio with the ability to change the size and the possibility of rotating the object by rotating plate, in which the object placed on it - if required, to multiple views (or the use of a robotic arm with a capacity of motion in all directions with a plate, consisting of 3 fixed square plates) (top, front and left or right) and a rotating plate (or a movable arm with a plate instead of 3 plates) is formed and the center of each plate includes a camera and 4 light sources and a laser pointer.

[Claim 3] According to claim 2, each camera has a laser pointer used to align

objects.

[Claim 4] According to claim 2, a resizable studio has 3 plates, or a robotic arm or a rotating plate.

[Claim 5] According to claim 2, in the first step, imaging of any required view is done without the presence of an object and a raw image will be created as a background image for each view.

[Claim 6] According to claim 2, in the next step, in each direction, taking into

account the existence of four different light sources, four images of the object are taken and, to apply texture on the final three-dimensional object in each view, an image will be taken using all four light sources at the same time.

[Claim 7] According to claims 5 and 6, after the end of imaging for each view, 6

images of the desired object and its background are created.

[Claim 8] According to claim 7, after imaging using software designed, images will be entered into the image processing stage using GPU processing, and the object boundary from other parts of the image will be identified using the SSD algorithm.

[Claim 9] According to claim 8, to remove the background image using

computations of different pixels in two images, by deleting the background image data of view without presence of an object, the image of the object will be separately prepared pure and without the background for each view.

[Claim 10] According to claim 9, the images are purified using different visual

machine techniques, and image processing for each direction of 4 RGB-D Pixel Matrix is calculated.

[Claim 11 ] According to claim 10, by combining four matrices and the amount of light reflection from 4 different angles, the normal vector is extracted for each pixel. Then, a two-dimensional profile on each plate of view with its normal vector is created for each pixel.

[Claim 12] According to claim 11 , based on the number of pixel profiles, the data in the coordinate plates for each view of the common pixels is compared to the normal vector, and, by combining the different views of the pixels of the images, a three-dimensional matrix will be constructed in dimensions of three directions X, Y, Z in normal and negative normal vectors.

[Claim 13] According to claim 12, matrix built using API Blender, after the calculation using the Blender kernel features of a fully automated Wavefront file, is rendered in 3 different views without requiring the user in raw form without material.

[Claim 14] According to claim 13, in order to obtain a suitable file map as a 3D file material, using the DCGAN neural network, a texture-specific images in the image on each one of three views after converting to map format alongside the 3D file format is entered the network and finally a good map file is obtained.

[Claim 15] According to claim 14, the final 3D file and the material are finally

combined using the Blender API as completely automated form without the need for the user, and an editable three-dimensional file will be extracted.