Processing

Please wait...

Settings

Settings

Goto Application

1. WO2020196985 - APPARATUS AND METHOD FOR VIDEO ACTION RECOGNITION AND ACTION SECTION DETECTION

Publication Number WO/2020/196985
Publication Date 01.10.2020
International Application No. PCT/KR2019/004798
International Filing Date 22.04.2019
IPC
G06K 9/00 2006.01
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
9Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
H04N 21/845 2011.01
HELECTRICITY
04ELECTRIC COMMUNICATION TECHNIQUE
NPICTORIAL COMMUNICATION, e.g. TELEVISION
21Selective content distribution, e.g. interactive television or video on demand
80Generation or processing of content or additional data by content creator independently of the distribution process; Content per se
83Generation or processing of protective or descriptive data associated with content; Content structuring
845Structuring of content, e.g. decomposing content into time segments
G06N 3/08 2006.01
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
08Learning methods
G11B 27/36 2006.01
GPHYSICS
11INFORMATION STORAGE
BINFORMATION STORAGE BASED ON RELATIVE MOVEMENT BETWEEN RECORD CARRIER AND TRANSDUCER
27Editing; Indexing; Addressing; Timing or synchronising; Monitoring; Measuring tape travel
36Monitoring, i.e. supervising the progress of recording or reproducing
G06K 9/42 2006.01
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
9Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
36Image preprocessing, i.e. processing the image information without deciding about the identity of the image
42Normalisation of the pattern dimensions
Applicants
  • 연세대학교 산학협력단 INDUSTRY-ACADEMIC COOPERATION FOUNDATION, YONSEI UNIVERSITY [KR]/[KR]
Inventors
  • 손광훈 SOHN, Kwang Hoon
  • 박정인 PARK, Jung In
Agents
  • 민영준 MIN, Young Joon
Priority Data
10-2019-003483227.03.2019KR
Publication Language Korean (KO)
Filing Language Korean (KO)
Designated States
Title
(EN) APPARATUS AND METHOD FOR VIDEO ACTION RECOGNITION AND ACTION SECTION DETECTION
(FR) APPAREIL ET PROCÉDÉ DE RECONNAISSANCE D'ACTION VIDÉO ET DE DÉTECTION DE SECTION D'ACTION
(KO) 비디오 행동 인식 및 행동 구간 탐지 장치 및 방법
Abstract
(EN)
The present invention may provide an apparatus and method for video action recognition and action section detection, which can perform temporal action localization on a video by being trained using a training video in which only a simple action label is annotated, thereby reducing temporal and cost burdens for obtaining the training video, and can recognize an accurate temporal location of an action with temporal consistency by extracting, from the video, feature maps according to segments to analyze action reliability according to the segments and a semantic similarity between the segments regarding a same action, and applying a weight to the action reliability according to the segments on the basis of the semantic similarity between the segments.
(FR)
La présente invention concerne un appareil et un procédé de reconnaissance d'action vidéo et de détection de section d'action, capable réaliser une localisation d'action temporelle sur une vidéo en étant entraîné à l'aide d'une vidéo d'entraînement dans laquelle seule une simple étiquette d'action est annotée, ce qui réduit les charges temporelles et financières pour obtenir la vidéo d'entraînement et capable de reconnaître une localisation temporelle précise d'une action avec une cohérence temporelle en extrayant, à partir de la vidéo, des cartes de caractéristiques selon des segments pour analyser une fiabilité d'action selon les segments et une similitude sémantique entre les segments concernant une même action et en appliquant une pondération à la fiabilité d'action selon les segments sur la base de la similitude sémantique entre les segments.
(KO)
본 발명은 간단한 행동 레이블만이 주석된 학습용 비디오를 이용하여 학습되어 비디오에 대한 시간적 행동 로컬라이제이션을 수행할 수 있어, 학습용 비디오를 획득하기 위한 시간적 비용적 부담을 경감할 수 있으며, 비디오에서 세그먼트별 특징맵을 추출하여 세그먼트별 행동 신뢰도와 동일 행동에 대한 세그먼트 간 시멘틱 유사성을 분석하여 세그먼트별 행동 신뢰도에 세그먼트 간 시멘틱 유사성을 기반으로 가중치를 적용함으로써, 시간적 일관성을 갖고 행동의 정확한 시간적 위치를 인식할 수 있는 비디오 행동 인식 및 행동 구간 탐지 장치 및 방법을 제공할 수 있다.
Latest bibliographic data on file with the International Bureau