WO2022045519A1 - Dispositif et procédé de sélection de modèle d'optimisation - Google Patents

Dispositif et procédé de sélection de modèle d'optimisation Download PDF

Info

Publication number
WO2022045519A1
WO2022045519A1 PCT/KR2021/005179 KR2021005179W WO2022045519A1 WO 2022045519 A1 WO2022045519 A1 WO 2022045519A1 KR 2021005179 W KR2021005179 W KR 2021005179W WO 2022045519 A1 WO2022045519 A1 WO 2022045519A1
Authority
WO
WIPO (PCT)
Prior art keywords
recognition
object recognition
modeling program
rate
rates
Prior art date
Application number
PCT/KR2021/005179
Other languages
English (en)
Korean (ko)
Inventor
이진석
이승원
김범진
김형복
윤석원
Original Assignee
주식회사 테스트웍스
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 주식회사 테스트웍스 filed Critical 주식회사 테스트웍스
Publication of WO2022045519A1 publication Critical patent/WO2022045519A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/96—Management of image or video recognition tasks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/174—Segmentation; Edge detection involving the use of two or more images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/10—Image acquisition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning

Definitions

  • the present invention relates to an apparatus and method for selecting an optimization model, and more particularly, when learning using a new object recognition modeling program, selecting an optimization model for learning by selecting an object model optimized according to processing speed and recognition rate It relates to an apparatus and method.
  • the motion history algorithm is an algorithm for estimating the motion of an object by binarizing values above a threshold in each result image through the difference operation of adjacent frames, and then overlaying the result images for a short time.
  • objects can be recognized well regardless of time because objects are recognized through frame-to-frame object matching based on feature points.
  • model images of the object to be recognized must be input in advance, there is a very limited problem in detecting a large number of unspecified objects.
  • Mean-shift using the histogram back-projection technique measures the similarity in the input image using a hue channel histogram, and as a result, the pixel value of the image is changed to a probability value, and the mean-shift is performed on the distribution of probability values.
  • FIG. 1 is a block diagram schematically showing the configuration of an object recognition apparatus according to the prior art.
  • the object recognition apparatus 10 of FIG. 1 (a) includes an image input unit 11 to which an image is input, a time-priority modeling control unit 12 for recognizing an object from an image, and a method for displaying a recognition result
  • the display unit 13 may be configured.
  • the object recognition apparatus 10 of FIG. 1(a) has the advantage of fast object recognition in the object recognition modeling program, which has a small amount of computation and a fast recognition time, but it has a problem that it is difficult to utilize it as it is in an object recognition service where the recognition rate is important because the recognition rate is low. .
  • the object recognition apparatus 20 of FIG. 1(b) includes an image input unit 21 to which an image is input, a performance-priority modeling control unit 22 for recognizing an object from an image, and a display unit 23 for displaying the recognition result ) can be composed of
  • the object recognition apparatus 20 of FIG. 1B has a high recognition rate but a large amount of computation, so it takes a lot of time to calculate the recognition rate, so it is difficult to use it as it is in an object recognition service where real-time recognition is important.
  • an optimized object recognition modeling program capable of improving the real-time recognition rate while maintaining or increasing the recognition rate of an object above a certain standard is needed.
  • an object of the present invention is to provide an optimization model selection apparatus and method for selecting an object model optimized according to processing speed and recognition rate when learning using a new object recognition modeling program and performing learning do it with
  • an embodiment of the present invention is an apparatus for selecting an optimization model, by inputting a data set composed of a plurality of images into a plurality of object recognition modeling programs, respectively, to determine the recognition rate of an individual object recognition modeling program
  • the calculated recognition rates are the same or the difference between the calculated recognition rates is satisfied without exceeding a preset reference range, selecting an object recognition modeling program with the fastest calculation speed of the recognition rate to perform object recognition characterized.
  • the plurality of object recognition modeling programs according to the embodiment are characterized in that they are a time-priority object recognition modeling program giving priority to time when calculating a recognition rate, and a performance-priority object recognition modeling program giving priority to a recognition rate.
  • the selection device recalculates the recognition rate by inputting a data set into a plurality of object recognition modeling programs at regular intervals, and the recalculated recognition rate is the same or a difference between the recalculated recognition rates is preset. If the range is satisfied without exceeding the range, the currently set time-priority object recognition modeling program is maintained, and if the calculated recognition rate does not satisfy the reference range, it is changed to the performance-priority object recognition modeling program.
  • the embodiment includes an image input unit for inputting a data set; and inputting the data set into a plurality of object recognition modeling programs, respectively, to calculate a recognition rate of an individual object recognition modeling program, but based on the calculated recognition rate, the recognition rate is the same or a difference between the calculated recognition rates exceeds a preset reference range It is characterized in that it comprises a modeling control unit for performing object recognition by selecting a time-priority object recognition modeling program that gives priority to time when calculating the recognition rate, if satisfied.
  • the modeling control unit recalculates the recognition rates of the plurality of object recognition modeling programs at regular intervals, but the recalculated recognition rates are the same or the difference between the recalculated recognition rates does not exceed a preset reference range If satisfied, the currently set time-priority object recognition modeling program is maintained, and if the difference between the recalculated recognition rates exceeds a reference range and is not satisfied, it is characterized by changing to a performance-priority object recognition modeling program that prioritizes the recognition rate .
  • an embodiment of the present invention provides a method for selecting an optimization model, a) the selection device inputs a data set composed of a plurality of images into a plurality of object recognition modeling programs, respectively, and the recognition rate of the individual object recognition modeling program calculating ; and b) if the calculated recognition rates are the same or the difference between the calculated recognition rates does not exceed a preset reference range and is satisfied, the selection device selects the object recognition modeling program with the fastest calculating speed of the recognition rate and performs object recognition including;
  • the embodiment includes the steps of: c) recalculating, by the selection device, a recognition rate by inputting a data set into a plurality of object recognition modeling programs at regular intervals; and d) if the recalculated recognition rates are the same or the difference between the recalculated recognition rates is satisfied without exceeding a preset reference range, the selection device maintains the currently set object recognition modeling program, and between the recalculated recognition rates If the difference exceeds the reference range and is not satisfied, it is characterized by changing to another object recognition modeling program.
  • the plurality of object recognition modeling programs according to the embodiment are characterized in that they are a time-priority object recognition modeling program giving priority to time when calculating a recognition rate, and a performance-priority object recognition modeling program giving priority to a recognition rate.
  • the present invention has the advantage that, when learning using a new object recognition modeling program, an object model optimized according to the processing speed and recognition rate can be selected for learning.
  • the present invention has the advantage of maintaining an appropriate recognition rate and improving current efficiency because two object recognition modeling programs operate by replacing them periodically according to the recognition rate.
  • FIG. 1 is a block diagram showing the configuration of an object recognition modeling programming system according to the prior art.
  • FIG. 2 is a block diagram showing the configuration of an optimization model selection apparatus according to an embodiment of the present invention.
  • FIG. 3 is a flowchart illustrating a method for selecting an optimization model according to an embodiment of the present invention.
  • FIG. 4 is another flowchart illustrating a method for selecting an optimization model according to the embodiment of FIG. 3;
  • ... unit means a unit that processes at least one function or operation, which may be divided into hardware, software, or a combination of the two.
  • the term "at least one" is defined as a term including the singular and the plural, and even if the term at least one does not exist, each element may exist in the singular or plural, and may mean the singular or plural. will be self-evident.
  • FIG. 2 is a block diagram illustrating the configuration of an apparatus for selecting an optimization model according to an embodiment of the present invention.
  • the selection apparatus 100 inputs a data set composed of a plurality of images into a plurality of object recognition modeling programs, respectively, and the recognition rate of the individual object recognition modeling programs to calculate
  • the selection device 100 selects an object recognition modeling program having the fastest calculation speed of the recognition rate and performs object recognition.
  • the selection device 100 may recalculate a recognition rate by inputting a data set into a plurality of object recognition modeling programs at regular intervals, and maintain or change the selected object recognition modeling program according to the result of the recalculated recognition rate, To this end, it is configured to include an image input unit 110 , a modeling control unit 120 , and a database 130 .
  • the plurality of object recognition modeling programs according to the embodiment may be composed of a time-priority object recognition modeling program giving priority to time when calculating a recognition rate, and a performance-priority object recognition modeling program giving priority to a recognition rate.
  • the image input unit 110 receives an image to be recognized, and receives an image stored in an internal storage device or an external storage device, or outputs it from a digital photographing device such as a CCD sensor, a CMOS sensor, or a camera having a photoelectric conversion means. image information can be received.
  • a digital photographing device such as a CCD sensor, a CMOS sensor, or a camera having a photoelectric conversion means. image information can be received.
  • the modeling control unit 120 inputs the data sets input from the image input unit 110 into a plurality of object recognition modeling programs, respectively, and calculates a recognition rate of each object recognition modeling program.
  • the modeling control unit 120 may include a first modeling unit 121 and a second modeling unit 122 , and the first modeling unit 121 includes a time in which time is prioritized when calculating the recognition rate.
  • an object recognition modeling program is installed to perform object recognition.
  • a performance-priority object recognition modeling program that prioritizes the recognition rate is installed in the second modeling unit 122 to perform object recognition.
  • the individual object recognition modeling programs installed in the first and second modeling units 121 and 122 perform object recognition using one or more of a machine learning program and a deep learning program on the input image, and display the object recognition result. It can be provided by calculating the recognition rate.
  • the modeling control unit 120 determines whether the recognition rates of the time-priority object recognition modeling program and the performance-priority object recognition modeling program are the same or are satisfied without exceeding a preset reference range based on the recognition rate calculated from the individual object recognition modeling program. judge
  • the modeling control unit 120 selects a time-priority object recognition modeling program that prioritizes time to perform object recognition and let it learn
  • the modeling control unit 120 may recalculate the recognition rates of all object recognition modeling programs at regular intervals while the time-priority object recognition modeling program is selected to perform and learn object recognition.
  • the modeling control unit 120 recalculates the recognition rates of all object recognition modeling programs after learning a data set of a certain time or a certain ratio (image processing amount, etc.).
  • the modeling control unit 120 determines that the currently set time-priority object recognition modeling program performs object recognition and learning. By keeping it possible, real-time performance can be improved.
  • the modeling control unit 120 changes to a performance-priority object recognition modeling program that prioritizes the recognition rate in order to improve the recognition rate, so that object recognition and learning are performed. By doing so, object recognition performance can be maintained.
  • the database 130 is a configuration for storing a machine learning program and a deep learning program, and the database 130 is configured as a storage medium physically included in the selection device 100 , but is separated from the selection device 100 . It may be installed in a remote location and may consist of a storage medium connected through a network.
  • the following describes a method for selecting an optimization model according to an embodiment of the present invention.
  • FIG. 3 is a flowchart illustrating a method for selecting an optimization model according to an embodiment of the present invention
  • FIG. 4 is another flowchart illustrating a method for selecting an optimization model according to the embodiment of FIG. 3 .
  • the selection device 100 receives a data set composed of a plurality of images through the image input unit 110 . (S100).
  • the data sets input in step S100 are respectively input to a plurality of object recognition modeling programs stored in advance by the selection device 100, and the result of object recognition by the individual object recognition modeling program for the input data sets is calculated as a recognition rate. Compare (S110).
  • the individual object recognition modeling program may be composed of a time-priority object recognition modeling program giving priority to time when calculating a recognition rate, and a performance-priority object recognition modeling program giving priority to a recognition rate.
  • step S110 It is determined whether the recognition rates of the individual object recognition modeling programs calculated in step S110 are the same or whether a difference between the calculated recognition rates is satisfied without exceeding a preset reference range (S120).
  • the selection device 100 selects the object recognition modeling program that has the fastest calculation speed of the recognition rate, that is, the time-priority object recognition modeling program that gives priority to time when calculating the recognition rate. It enables execution and learning (S130).
  • the selection device 100 performs object recognition and learning using the time-priority object recognition modeling program set in step S130, and inputs a data set into all object recognition modeling programs installed at regular intervals for intermediate evaluation.
  • learning is performed for each individual object recognition modeling program (S140), and the recognition rate is recalculated for each individual object recognition modeling program to determine the difference in the recognition rate (S150).
  • the selection device 100 is a currently set object recognition modeling program, that is, a time-priority object
  • the recognition modeling program is maintained, and object recognition is performed and learned based on the time-priority object recognition modeling program (S160).
  • step S150 if the recalculated recognition rate exceeds the reference range and is not satisfied, the selection device 100 recognizes another object recognition modeling program, that is, a performance-priority object recognition that takes a long time but prioritizes the recognition rate After changing to the modeling program, object recognition is performed and learning is performed (S200).
  • object recognition modeling program that is, a performance-priority object recognition that takes a long time but prioritizes the recognition rate
  • the selection device 100 performs object recognition and learning using the performance-priority object recognition modeling program set in step S200, and inputs a data set into all object recognition modeling programs installed at regular intervals for intermediate evaluation.
  • learning is performed for each individual object recognition modeling program (S210), and the recognition rate is recalculated for each individual object recognition modeling program to determine the difference in the recognition rate (S220).
  • step S220 if there is a difference between the recalculated recognition rates exceeding a preset reference range, the selection device 100 maintains the currently set object recognition modeling program, that is, the performance-priority object recognition modeling program, and Perform and learn object recognition based on the performance-priority object recognition modeling program (S230).
  • the selection device 100 executes the currently set object recognition modeling program, that is, the performance-priority object recognition modeling program for time. First, it is changed to an object recognition program, and object recognition is performed and learned based on the changed time-priority object recognition modeling program (S231).

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

L'invention divulgue un dispositif et un procédé de sélection de modèle d'optimisation. Lors de l'apprentissage à l'aide d'un nouveau programme de modélisation de reconnaissance d'objet, la présente invention peut effectuer un apprentissage par sélection d'un modèle d'objet qui est optimisé en fonction d'une vitesse de traitement et d'un taux de reconnaissance. Le dispositif de sélection de modèle d'optimisation est un dispositif de sélection (100) qui entre un ensemble de données composé d'une pluralité d'images dans chacun d'une pluralité de programmes de modélisation de reconnaissance d'objet pour calculer un taux de reconnaissance d'un programme de modélisation de reconnaissance d'objet individuel et, lorsque le taux de reconnaissance calculé est identique ou qu'une différence entre les taux de reconnaissance calculés est satisfaite sans dépasser une plage de référence prédéfinie, sélectionne un programme de modélisation de reconnaissance d'objet ayant la vitesse de calcul la plus rapide du taux de reconnaissance pour effectuer une reconnaissance d'objet.
PCT/KR2021/005179 2020-08-24 2021-04-23 Dispositif et procédé de sélection de modèle d'optimisation WO2022045519A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
KR10-2020-0106128 2020-08-24
KR1020200106128A KR102440073B1 (ko) 2020-08-24 2020-08-24 최적화 모델 선택 장치 및 방법

Publications (1)

Publication Number Publication Date
WO2022045519A1 true WO2022045519A1 (fr) 2022-03-03

Family

ID=80442750

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/KR2021/005179 WO2022045519A1 (fr) 2020-08-24 2021-04-23 Dispositif et procédé de sélection de modèle d'optimisation

Country Status (2)

Country Link
KR (1) KR102440073B1 (fr)
WO (1) WO2022045519A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116129731A (zh) * 2022-12-29 2023-05-16 北京布局未来教育科技有限公司 人工智能模拟教学系统与方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20170008251A (ko) * 2014-05-13 2017-01-23 도쿄엘렉트론가부시키가이샤 제조 프로세스를 모델링 및/또는 분석하기 위한 시스템 및 방법
KR101827108B1 (ko) * 2016-05-04 2018-02-07 두산중공업 주식회사 플랜트 이상 감지 학습 시스템 및 방법
KR20190050516A (ko) * 2017-11-03 2019-05-13 삼성전자주식회사 우선 순위에 기반하여 이미지를 처리하는 전자 장치 및 그 동작 방법
KR20190094133A (ko) * 2019-04-16 2019-08-12 엘지전자 주식회사 객체를 인식하는 인공 지능 장치 및 그 방법
KR102074073B1 (ko) * 2018-06-29 2020-03-02 한국교통대학교산학협력단 차량 인식 방법 및 이를 이용하는 장치

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9336770B2 (en) * 2013-08-13 2016-05-10 Mitsubishi Electric Corporation Pattern recognition apparatus for creating multiple systems and combining the multiple systems to improve recognition performance and pattern recognition method
JP7196542B2 (ja) * 2018-11-05 2022-12-27 株式会社リコー 学習装置および学習方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20170008251A (ko) * 2014-05-13 2017-01-23 도쿄엘렉트론가부시키가이샤 제조 프로세스를 모델링 및/또는 분석하기 위한 시스템 및 방법
KR101827108B1 (ko) * 2016-05-04 2018-02-07 두산중공업 주식회사 플랜트 이상 감지 학습 시스템 및 방법
KR20190050516A (ko) * 2017-11-03 2019-05-13 삼성전자주식회사 우선 순위에 기반하여 이미지를 처리하는 전자 장치 및 그 동작 방법
KR102074073B1 (ko) * 2018-06-29 2020-03-02 한국교통대학교산학협력단 차량 인식 방법 및 이를 이용하는 장치
KR20190094133A (ko) * 2019-04-16 2019-08-12 엘지전자 주식회사 객체를 인식하는 인공 지능 장치 및 그 방법

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116129731A (zh) * 2022-12-29 2023-05-16 北京布局未来教育科技有限公司 人工智能模拟教学系统与方法
CN116129731B (zh) * 2022-12-29 2023-09-15 北京布局未来教育科技有限公司 人工智能模拟教学系统与方法

Also Published As

Publication number Publication date
KR20220025435A (ko) 2022-03-03
KR102440073B1 (ko) 2022-09-05

Similar Documents

Publication Publication Date Title
WO2017213398A1 (fr) Modèle d'apprentissage pour détection de région faciale saillante
WO2019088462A1 (fr) Système et procédé pour générer un modèle d'estimation de pression artérielle, et système et procédé d'estimation de pression artérielle
KR101423916B1 (ko) 복수의 얼굴 인식 방법 및 장치
WO2016163755A1 (fr) Procédé et appareil de reconnaissance faciale basée sur une mesure de la qualité
WO2014069822A1 (fr) Appareil et procédé de reconnaissance de visage
WO2020027607A1 (fr) Dispositif de détection d'objets et procédé de commande
WO2015182904A1 (fr) Appareil d'étude de zone d'intérêt et procédé de détection d'objet d'intérêt
WO2018143486A1 (fr) Procédé de fourniture de contenu utilisant un système de modularisation pour analyse d'apprentissage profond
WO2010041836A2 (fr) Procédé de détection d'une zone de couleur peau à l'aide d'un modèle de couleur de peau variable
WO2021100919A1 (fr) Procédé, programme et système pour déterminer si un comportement anormal se produit, sur la base d'une séquence de comportement
WO2018151503A2 (fr) Procédé et appareil destinés à la reconnaissance de gestes
WO2022045519A1 (fr) Dispositif et procédé de sélection de modèle d'optimisation
CN115065798B (zh) 一种基于大数据的视频分析监控系统
CN108052931A (zh) 一种车牌识别结果融合方法及装置
WO2022114653A1 (fr) Système et procédé de calcul de limite de données
WO2016064107A1 (fr) Procédé et appareil de lecture vidéo sur la base d'une caméra à fonctions de panoramique/d'inclinaison/de zoom
WO2021145502A1 (fr) Dispositif et procédé de reconnaissance faciale et de stress
WO2019112385A1 (fr) Procédé de codage d'informations temporelles de caractéristiques spécifiques à une trame de segment d'image en vue d'une reconnaissance de vidéo
WO2021206338A1 (fr) Procédé et dispositif de reconnaissance de caractères sur conteneur de fret
WO2024019337A1 (fr) Procédé et appareil d'amélioration de vidéo
WO2017222228A1 (fr) Procédé de reconnaissance de transition d'écran dans un contenu d'image, et serveur de commande correspondant
WO2014185691A1 (fr) Appareil et procédé d'extraction d'image à filigrane élevé à partir d'images photographiées en continu
WO2024101466A1 (fr) Appareil et procédé de suivi de personne disparue basé sur des attributs
WO2022019601A1 (fr) Extraction d'un point caractéristique d'un objet à partir d'une image ainsi que système et procédé de recherche d'image l'utilisant
WO2020101121A1 (fr) Procédé d'analyse d'image basée sur l'apprentissage profond, système et terminal portable

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: 21861821

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: 21861821

Country of ref document: EP

Kind code of ref document: A1