WO2022146166A1 - Plateforme d'instructions par étapes de réalité augmentée à des fins techniques - Google Patents

Plateforme d'instructions par étapes de réalité augmentée à des fins techniques Download PDF

Info

Publication number
WO2022146166A1
WO2022146166A1 PCT/RU2020/000785 RU2020000785W WO2022146166A1 WO 2022146166 A1 WO2022146166 A1 WO 2022146166A1 RU 2020000785 W RU2020000785 W RU 2020000785W WO 2022146166 A1 WO2022146166 A1 WO 2022146166A1
Authority
WO
WIPO (PCT)
Prior art keywords
instructions
model
equipment
image
module
Prior art date
Application number
PCT/RU2020/000785
Other languages
English (en)
Russian (ru)
Inventor
Дмитрий Анатольевич КУЗЬМЕНКО
Наталья Сергеевна ЛЕВЧЕНКО
Юрий Муратович НАБОКОВ
Егор Алексеевич НАРЫШКИН
Алексей Александрович ОСТРОВЕРХОВ
Тимофей Юрьевич САВИН
Иван Владимирович СОКОЛОВСКИЙ
Сергей Леонидович СОЛЯНИК
Юлия Петровна СОЛЯНИК
Original Assignee
Общество С Ограниченной Ответственностью "Спайдер Груп"
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Общество С Ограниченной Ответственностью "Спайдер Груп" filed Critical Общество С Ограниченной Ответственностью "Спайдер Груп"
Priority to PCT/RU2020/000785 priority Critical patent/WO2022146166A1/fr
Publication of WO2022146166A1 publication Critical patent/WO2022146166A1/fr

Links

Classifications

    • G—PHYSICS
    • G02—OPTICS
    • G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
    • G02B27/00—Optical systems or apparatus not provided for by any of the groups G02B1/00 - G02B26/00, G02B30/00
    • G02B27/01—Head-up displays
    • G—PHYSICS
    • G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
    • G09G—ARRANGEMENTS OR CIRCUITS FOR CONTROL OF INDICATING DEVICES USING STATIC MEANS TO PRESENT VARIABLE INFORMATION
    • G09G5/00—Control arrangements or circuits for visual indicators common to cathode-ray tube indicators and other visual indicators
    • G09G5/22—Control arrangements or circuits for visual indicators common to cathode-ray tube indicators and other visual indicators characterised by the display of characters or indicia using display control signals derived from coded signals representing the characters or indicia, e.g. with a character-code memory

Definitions

  • the invention relates to the field of computer analysis of images, in particular to the use of artificial intelligence (AI) to determine the types and models of objects of technology through the analysis of the image coming from the camera of the user's device, and the subsequent demonstration of instructions for the object of technology using augmented reality (AR) technology through visual linking hints to a real object of technology.
  • AI artificial intelligence
  • AR augmented reality
  • the presented solution can be used, at least in domestic conditions, when interacting with home appliances, as well as in industrial production when interacting with devices, including those requiring detailed study of regulatory documents.
  • US Pat. No. 9,324,229 B2, April 26, 2016, describes a head-mounted display by which a user looking at a suitably marked instrument panel can be provided with an overlay image with instructions for the operation and maintenance of the equipment directly superimposed on the instrument panel.
  • the head-mounted display provides the user with a magnified view of the object being viewed; a tracking mechanism, such as a camera, repeatedly determines the position and orientation of the head-mounted display relative to the object being viewed; and the computer system provides information for the expanded view and repeatedly updates the expanded view of the object being viewed based on the determined position and orientation of the display.
  • the tracking mechanism determines its position using one or more markers or beacons on the viewed object. Markers can be active or passive, including light-emitting diodes (LEDs) that emit invisible light.
  • LEDs light-emitting diodes
  • Patent KR 102171691 B1, 10/29/2020 describes a method and system for servicing a 3D printer based on augmented reality, which includes the steps of recognizing an auxiliary object of the 3D printer based on previously stored data when viewing the 3D printer through the camera of the user terminal; the formation of a virtual frame, including an auxiliary object; return of information about the camera coordinates for a certain point of the virtual frame; and outputting computer graphics with information about the auxiliary object recognized based on the coordinate information, and displaying it on the camera screen of the user terminal and captured on the camera screen of the user terminal based on augmented reality.
  • the closest analogue of the claimed invention is the technical solution disclosed in the application WO 2015125066 A1, 08/27/2015.
  • a system is described to facilitate the maintenance of equipment during field work by mobile technicians.
  • the system includes a server configured to store hardware configurations and service protocols; an equipment maintenance logger that communicates with a server, and a variety of smart glasses worn by field technicians, the system provides the field technician with interactive instructions that are displayed on the smart glass to help with field work.
  • this solution lacks the ability to recognize the type and model of equipment using computer technology, and the field technician determines the type and model of equipment manually on site, or performs maintenance on equipment whose type and model is already known and recognition is not required.
  • the technical problem to be solved by the claimed invention is to create a comprehensive scalable solution for determining the types and models of various objects of technology through image analysis, and then demonstrating instructions for the recognized model of technology using augmented reality (AR) technology, which, in addition to tools for issuing instructions for step-by-step manipulation also contains tools for creating content and determining the type and model of an object, which allows you to exclude the assistance of a user by a technical specialist from the service process, and also contains an AI-based tracking algorithm that does not require the creation of a 3D model and, at the same time, allows work with three-dimensional objects, track and bind marks from different sides of the object. Also, the problem to be solved is the qualitative identification of the model and the implementation of high-quality tracking (binding labels).
  • AR augmented reality
  • the technical result of the claimed invention is the implementation of operations for visualizing instructions for various models of equipment in the AR mode more quickly and efficiently, reducing the time it takes to search for and displaying instructions to the end user for various models of equipment, excluding from the maintenance and operation of the technique of assisting the end user by a technician, which in ultimately leads to the simplification of maintenance and operation, the acceleration of maintenance work, the reduction of errors to zero in the performance of maintenance operations and the operation of equipment by the end user, and also contributes to an increase in the life of the equipment.
  • the specified technical result is achieved due to the fact that in the method of demonstrating instructions for a technical object using augmented reality, the following steps are carried out: capturing an image of a technical object using the end user device; recognition of the type and model of the equipment object based on the current image frame; transferring to the end user device a data set for the recognized model of equipment, containing at least a reference image, coordinates of interface elements, a list of instructions, detailed information for instructions; extracting at least key points and descriptors for the reference image and the current frame, matching key points by descriptors; building a tomography matrix to obtain the coordinates of the interface elements on the current frame, projecting the coordinates of the interface elements onto the current frame; displaying instructions for the recognized model of the vehicle object on the current frame on the end user device.
  • the system for demonstrating instructions for a technical object using augmented reality comprises: an end user device containing an image capture means and a means for displaying information to the user; classification module; dataset of training images; tracking module; updated catalog of instructions; which carry out the capture of the image of the object of technology using the image capture means; based on the current image frame, the type and model of the technical object are recognized by means of the classification module; projecting the coordinates of the interface elements on the reference image of the recognized model of the technical object onto the current frame by means of the tracking module; display instructions for the recognized model of the vehicle object using augmented reality, where instruction step labels are linked to the real vehicle object displayed on the end user's device.
  • the end user device can be a smartphone, tablet, or AR glasses.
  • the image capture medium can be the built-in camera.
  • an object of equipment can be an object of household, professional, industrial equipment.
  • the classification module may include a technique type determination module and a technique model determination module.
  • the system can additionally determine the brand of equipment.
  • the module for determining the type of equipment and the tracking module can be implemented on the end user device, the module for determining the model of equipment and the catalog of instructions can be implemented on the server.
  • the system can store images of equipment objects for training the module for determining the model of equipment in a dataset of training images.
  • the system can recognize the model of the object of technology using neural networks.
  • the system can carry out initial training of neural networks based on 10 or more different photographs of a technical object.
  • data about the object of technology can come from various sources, at least from the internal product team and from end users.
  • the system can carry out, by means of the tracking module, at least tracking through the detection and short-term tracking of the technical object.
  • the system may further comprise an instruction creation interface, wherein the instruction catalog is replenished using the instruction creation interface.
  • the claimed invention provides a system in which it is possible to create and visualize instructions for a scalable catalog of machinery and equipment.
  • the end user simply points the phone at the equipment, and the system itself determines the type, brand and model of the equipment; without forcing the user to think about how to determine the model or, if it is impossible to determine the model of equipment by the user himself, seek the help of a technical specialist.
  • the system immediately gives the user a list of instructions for it, otherwise it offers a manual search in the catalog. After the user selects an instruction, the system displays instructions using AR technology, where instruction step labels are tied to a real object.
  • the user is shown instructions on a mobile phone using AR technology.
  • the claimed invention has its own tracking algorithm based on AI technology. This algorithm does not require the creation of a 3D model, unlike many other solutions, and, at the same time, allows you to work with volumetric objects, track and bind marks from different sides. object. Initially, the side of the object is identified and an arrow (or text hint) is shown to the user, which tells the user exactly where to point the camera (to which side of the object).
  • the system includes the following components:
  • the solution uses pre-trained neural networks, while each time receiving data from the user about the successful or unsuccessful definition of the model, additional training of networks can be carried out. This process allows you to get better results for the identification of objects.
  • the very process of initial training of neural networks takes place on the basis of at least 10 different photographs of the object.
  • the main goal in terms of model identification is to construct a neural network capable of learning from just a few photos (Single Shot Learning). That is, in order to determine a model of equipment previously unknown to the network, only a few photographs of equipment are needed, instead of hundreds or thousands of photos.
  • the collection of data about the models required for training, which are part of the neural network solution, is distributed.
  • the data comes from various sources: from the internal product team and from the end users of the solution. This allows the solution to develop faster in two directions: extensive (parallel collection of information about several models at once) and intensive (photos of the same model are received from different participants, but in different quality, lighting and camera angle) without the involvement of a highly specialized specialist.
  • the claimed invention also includes an algorithm for creating instructions, which involves several steps: marking all the elements of the object's interface on one or more photos, creating instructions using a simple visual editor.
  • the instruction creator simply specifies the sequence of object controls and adds text comments as needed.
  • the camera of the end user's device is pointed at the object of technology, a frame is captured on the user's mobile device, classified by type of equipment, the mobile device transmits the video stream in the form of frames to the server, the video server receives the video stream and parses it into separate frames, the frames from the video stream are stored on the Redis server .
  • the neural network algorithm recognizes the brand and specific model of equipment.
  • the neural network is implemented as a separate service (Daemon), which can be located both on the server itself and on separate recognition servers specially designed for this purpose.
  • the number of neural network servers is not limited in number - they can connect to the system as needed, ensuring the horizontal growth of the solution.
  • Each neural network server receives new frames from the Redis server for processing, removing the frame from the queue, with each frame marked with a label (stamp). After the object recognition is completed, the neural network server returns the result of its work to the Redis server with the same label (stamp) that was assigned to a particular frame.
  • the list of available instructions for the recognized model is displayed on the end user's device, the user initiates a transition to a specific instruction, the data is received for initialization, the instruction step is determined.
  • the tracking module looks for objects associated with the instruction step on frames and keeps track of them.
  • the tracking module first searches for the descriptors of the frame received from the client, then compares it with the descriptor corresponding to the selected step, searches for the homography matrix, and maps the marker image objects to the received frame.
  • Marker image objects are selected rectangular fragments on the image.
  • Descriptors are a deep description of the frame points found using superpoints (a pre-trained neural network that extracts the key points of the given frame, and they are the same for the marker and the incoming frame). That is, a frame is input, the output is a set of "special" (according to the neural network) points with descriptions extracted by the neural network.
  • the result of the work of the neural network is the coordinates of the desired object on the frame.
  • the model is defined on the server, and the object is tracked on the mobile device.
  • Fig. 1 - illustrates the decision context diagram
  • Fig. 2 the algorithm of the tracking module
  • FIG. 4 is a general diagram of a computing device.
  • Figure 1 shows the overall design of the solution, consisting of 4 main components:
  • the end user device (1) can be a smartphone, tablet or AR glasses.
  • the main requirement for the device is the ability to capture a frame using the built-in camera, as well as the subsequent output of information to the user.
  • the component is designed to capture an image and display instructions for the model.
  • the server part (4) includes a module for determining the model of equipment (5) and a catalog of instructions (6).
  • the module for determining the model of technology (5) is implemented on the basis of the algorithm for classifying and clarifying the coordinates of objects and the algorithm for generating applicants.
  • the instruction catalog (6) is stored in the cloud.
  • the server part component (4) is designed to implement the model definition logic separately from the end user device.
  • the instruction creation interface (7) may be a web platform for facilitating the staff creation process.
  • An alternative to the web platform is the import of pre-prepared instructions by a script in csv / xml format.
  • the training image dataset (8) can be a cloud space or any other medium that stores vehicle images in JPG and/or PNG formats for training the vehicle model definition module (5).
  • Fig.1 The interaction of the main components in Fig.1 can be carried out using wired and/or wireless communication using REST and/or HTTP data transfer protocols.
  • FIG. 2 shows the algorithm of the tracking module.
  • the current algorithm is used only to set the object of interest at the beginning of work, change of the interface element or side, or when tracking is lost.
  • the tracking itself will be carried out using short-term tracking methods.
  • the algorithm consists of the following steps:
  • the input of the tracking module is a reference image in a single-channel format with a resolution of 640 by 480 pixels, for a certain model of equipment, the coordinates of the interface elements on this image and the current frame.
  • Feature points are matched by descriptors using a pre-trained SuperGlue neural network.
  • a homography matrix is built to obtain the coordinates of interface features of interest in the input image.
  • any two images of the same flat object in space are connected by homography. Having a set of points on the reference image and a set of points in the scene associated with it, it is possible to find a correspondence between them in the form of a homography matrix H using the RANSAC algorithm.
  • the algorithm evaluates homography for randomly selected points and does so until a sufficient match between coordinates is achieved.
  • a perspective matrix transformation of vectors is performed - multiplication of the homography matrix by the coordinates of points on the reference image. This operation allows you to find the desired coordinates on the frame.
  • the following are passed to the tracking module: an instruction, an instruction step, and a frame received from the user. Descriptors are extracted from the received frame and pre-prepared descriptors corresponding to the instruction step are loaded.
  • the marker descriptors and descriptors of the received frame are passed to the matcher, which matches the points of 2 descriptors and, at the output, looks for a homography matrix between images.
  • Descriptors are built according to the same principle both for marker images and for incoming frames from the user.
  • marked objects are displayed from the marker image to the submitted frame, drawn and served to the client.
  • mapping of the marker image into the given frame all objects of interest are also built and displayed.
  • the coordinates of the object are extracted already from the space of the user frame. These coordinates are transferred to the client, and already on the client side, rectangular objects are drawn according to the transferred coordinates.
  • the instruction output interface is rendered according to the step.
  • any convolutional artificial neural network is structurally divided into two parts.
  • the first one consists of convolutional and unifying layers and forms a feature matrix based on the original image.
  • the second part of the network is the classifier, which, having taken a set of features, produces a vector with probabilities for each class.
  • the definition of the brand of technology is carried out on the available limited set of images.
  • the best solution for classifying equipment by brand, taking into account the fact that almost all presented logos contain text, is to implement the module in two stages: text recognition on the frame and text comparison with logo templates. Training is continuous and is carried out for a large number of classes with a small sample size (1-10 photographs).
  • one or more photos of a vehicle model are uploaded, indicating its properties (type of vehicle, brand, model number).
  • the actor defines the area of the device marker and marks in it all the necessary controls and/or controls. For each of them, it specifies the type (for example, button, lever, switch, etc.) and how to interact with it (for example, press the button).
  • the actor saves the vehicle model with a set of controls in the catalog.
  • the model of equipment can have an identifier for linking to it the data received from the SCADA systems of industrial enterprises. This helps to visualize the data received from SCADA in relation to the control and / or control body.
  • the step of the instruction and the required control and/or control element are indicated.
  • the actor specifies a textual description of the instruction step when using the chain of blocks “action - control - value” from existing directories or by entering manually.
  • the previous step is repeated within a set of instruction steps.
  • the actor stores the instruction in the directory.
  • the camera of the end user device is aimed at the object.
  • the current frame is determined.
  • the probability vector of belonging to each of the given types of equipment is determined. Information about the frame and the probability vector is transmitted to the server side.
  • the neural network algorithm sequentially determines the brand and model of the equipment. If the vehicle model is not recognized, only the vehicle brand and category are reported to the end user (based on the probability vector). After determining the vehicle model, the server sends a set of data to the end user device: a reference image of the vehicle model, coordinates of its buttons, a list of instructions for this model.
  • the list of available instructions for the recognized model is displayed on the end user's device.
  • the user initiates a jump to a particular instruction.
  • markup means finding the coordinates of the rectangles that bound any part of the image that is significant for us at the current step, such a significant part of the image can be a button on a washing machine, kettle, and any other equipment.
  • the marker descriptors are passed to the marker tracking component for image detection.
  • the user determines the step of the instruction, then the process of matching the marker occurs when the camera is pointed at the object of interest to the user.
  • Information about the displayed step of the instruction is retrieved - it is the coordinates of the rectangle on the marker image (marked coordinates in the space of the marker image).
  • the marker image and the image submitted by the user are actually tensors (matrices) with the following dimensions - (3,640,480), which are further converted from RGB to Grayscale and get simply (1,640,480).
  • a homomorphism is constructed that maps the marker image to the space of the submitted frame, and through this homomorphism the frame coordinates are converted from the space of the marker to the space of the frame submitted by the user.
  • FIG. 4 shows an option for displaying an instruction step to the end user, for example, step 2/11 "Display” and step 5/11 "Socket".
  • FIG. 5 shows a general diagram of a computing device (N00) that provides the data processing necessary to implement the claimed solution.
  • a device contains components such as: one or more processors (N01), at least one memory (N02), storage media (N03), I/O interfaces (N04), I/O ( N05), networking tools (N06).
  • processors N01
  • memory N02
  • storage media N03
  • I/O interfaces N04
  • I/O N05
  • networking tools N06
  • the processor (N01) of the device performs the basic computing operations necessary for the operation of the device (N00) or the functionality of one or more of its components.
  • the processor (N01) executes the necessary machine-readable instructions contained in the main memory (N02).
  • Memory (N02), as a rule, is made in the form of RAM and contains the necessary software logic that provides the required functionality.
  • the data storage facility (N03) can be implemented in the form of HDD, SSD disks, raid array, network storage, flash memory, optical storage media (CD, DVD, MD, Blue-Ray disks), etc.
  • the tool (N03) allows you to perform long-term storage of various types of information.
  • Interfaces (N04) are standard means for connecting and working with the server part, for example, USB, RS232, RJ45, LPT, COM, HDMI, PS/2, Lightning, FireWire, etc.
  • the choice of interfaces (N04) depends on the specific version of the device (N00), which can be a personal computer, mainframe, server cluster, thin client, smartphone, laptop, etc.
  • the keyboard should be used as data I/O (N05) in any implementation of the system.
  • the keyboard hardware can be any known: it can be either a built-in keyboard used on a laptop or netbook, or a separate device connected to a desktop computer, server, or other computer device.
  • the connection can be either wired, in which the keyboard connection cable is connected to the PS / 2 or USB port located on the system unit of the desktop computer, or wireless, in which the keyboard exchanges data via a wireless communication channel, for example, a radio channel, with base station, which, in turn, is directly connected to the system unit, for example, to one of the USB ports.
  • I/O devices can also use: joystick, display (touchscreen), projector, touchpad, mouse, trackball, light pen, speakers, microphone, etc.
  • Means of network interaction are selected from devices that provide network reception and transmission of data, for example, an Ethernet card, WLAN / Wi-Fi module, Bluetooth module, BLE module, NFC module, IrDa, RFID module, GSM modem, etc.
  • N05 the organization of data exchange over a wired or wireless data transmission channel, for example, WAN, PAN, LAN (LAN), Intranet, Internet, WLAN, WMAN or GSM, 3G, 4G, 5G, is provided.
  • the device components (N00) are connected via a common data bus (N10).

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Optics & Photonics (AREA)
  • Computer Hardware Design (AREA)
  • Theoretical Computer Science (AREA)
  • User Interface Of Digital Computer (AREA)

Abstract

L'invention se rapporte au domaine de l'utilisation de l'intelligence artificielle (AI) pour déterminer des types et des modèles d'objets techniques. Le résultat technique de la présente invention consiste en la visualisation d'instructions pour différents modèles techniques en mode IA plus rapidement et avec une meilleure qualité, une diminution du temps de recherche et d'émission d'instructions, l'exclusion de l'assistance d'un spécialiste technique. Ce résultat test obtenu du fait que le procédé de démonstration des instructions pour un objet technique à l'aide de l'IA consiste à: saisir une image de l'objet; reconnaître le type et le modèle de l'objet technique; extraire et comparer des points particuliers; représenter les instructions pour le modèle d'objet reconnu.
PCT/RU2020/000785 2020-12-30 2020-12-30 Plateforme d'instructions par étapes de réalité augmentée à des fins techniques WO2022146166A1 (fr)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/RU2020/000785 WO2022146166A1 (fr) 2020-12-30 2020-12-30 Plateforme d'instructions par étapes de réalité augmentée à des fins techniques

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/RU2020/000785 WO2022146166A1 (fr) 2020-12-30 2020-12-30 Plateforme d'instructions par étapes de réalité augmentée à des fins techniques

Publications (1)

Publication Number Publication Date
WO2022146166A1 true WO2022146166A1 (fr) 2022-07-07

Family

ID=82260947

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/RU2020/000785 WO2022146166A1 (fr) 2020-12-30 2020-12-30 Plateforme d'instructions par étapes de réalité augmentée à des fins techniques

Country Status (1)

Country Link
WO (1) WO2022146166A1 (fr)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8814691B2 (en) * 2010-02-28 2014-08-26 Microsoft Corporation System and method for social networking gaming with an augmented reality
US8947456B2 (en) * 2012-03-22 2015-02-03 Empire Technology Development Llc Augmented reality process for sorting materials

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8814691B2 (en) * 2010-02-28 2014-08-26 Microsoft Corporation System and method for social networking gaming with an augmented reality
US8947456B2 (en) * 2012-03-22 2015-02-03 Empire Technology Development Llc Augmented reality process for sorting materials

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
ANONYMOUS: "How to Train YOLOv5 On a Custom Dataset", ROBOFLOW, 10 June 2020 (2020-06-10), pages 1 - 13, XP055953753, Retrieved from the Internet <URL:https://blog.roboflow.com/how-to-train-yolov5-on-a-custom-dataset/> [retrieved on 20220823] *
SINEM GUVEN KAYA ET AL., FINE-GRAINED VISUAL RECOGNITION FOR MOBILE AR TECHNICAL SUPPORT, 21 November 2020 (2020-11-21), Retrieved from the Internet <URL:https://web.archive.org/web/20201121150226> [retrieved on 20210907] *
ZHOU BING; GUVEN SINEM: "Fine-Grained Visual Recognition in Mobile Augmented Reality for Technical Support", IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, IEEE, USA, vol. 26, no. 12, 17 September 2020 (2020-09-17), USA, pages 3514 - 3523, XP011819358, ISSN: 1077-2626, DOI: 10.1109/TVCG.2020.3023635 *

Similar Documents

Publication Publication Date Title
US11321583B2 (en) Image annotating method and electronic device
Zimmermann et al. Learning to estimate 3d hand pose from single rgb images
JP6397144B2 (ja) 画像からの事業発見
CN109635621A (zh) 用于第一人称视角中基于深度学习识别手势的系统和方法
WO2019245768A1 (fr) Système de prédiction d&#39;emplacement de caractéristique d&#39;objet articulé
US11094079B2 (en) Determining a pose of an object from RGB-D images
WO2013175792A1 (fr) Système d&#39;estimation d&#39;attribut de personne et dispositif de génération de données d&#39;utilisation d&#39;apprentissage
WO2015135324A1 (fr) Procédé et terminal de tri d&#39;images
US11681409B2 (en) Systems and methods for augmented or mixed reality writing
CN111667005B (zh) 一种采用rgbd视觉传感的人体交互系统
JP7126586B2 (ja) 顔合成画像検出方法、顔合成画像検出装置、電子機器、記憶媒体及びコンピュータプログラム
CN110942511B (zh) 室内场景模型重建方法及装置
CN114782901B (zh) 基于视觉变动分析的沙盘投影方法、装置、设备及介质
CN110516707B (zh) 一种图像标注方法及其装置、存储介质
CN115008454A (zh) 一种基于多帧伪标签数据增强的机器人在线手眼标定方法
CN112328088B (zh) 图像的呈现方法和装置
KR20150022158A (ko) 기계제도 학습장치 및 학습방법
CN111638792A (zh) Ar效果的呈现方法、装置、计算机设备及存储介质
WO2022146166A1 (fr) Plateforme d&#39;instructions par étapes de réalité augmentée à des fins techniques
CN115661254A (zh) 一种多人姿态估计方法及相关装置
CN111124106A (zh) 一种追踪虚拟现实系统的方法
US11861899B2 (en) Systems and methods for augmented reality using web browsers
CN117173731B (zh) 一种模型训练的方法、图像处理的方法以及相关装置
JP2019185349A (ja) 検索装置、検索方法及びプログラム
CN114973294B (zh) 基于图文匹配方法、装置、设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20968107

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20968107

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 11.12.2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20968107

Country of ref document: EP

Kind code of ref document: A1