ETRI-Knowledge Sharing Plaform

ENGLISH

성과물

논문 검색
구분 SCI
연도 ~ 키워드

상세정보

학술대회 Behavior Recognition of a Person in a Daily Video Using Joint Position Information
Cited 3 time in scopus Download 1 time Share share facebook twitter linkedin kakaostory
저자
서정민, 유한주, 윤기민, 김현일, 최상일
발행일
201809
출처
Artificial Intelligence and Knowledge Engineering (AIKE) 2018: Workshops, pp.172-174
DOI
https://dx.doi.org/10.1109/AIKE.2018.00040
협약과제
18HS4600, (딥뷰-1세부) 실시간 대규모 영상 데이터 이해·예측을 위한 고성능 비주얼 디스커버리 플랫폼 개발, 박종열
초록
In this paper, we propose a method to recognize people's specific behaviors in images acquired in everyday life. The poses that occur when taking a specific action are represented using the joint coordinates of the body. To reduce the influence of the perspective effect on the locations in an image, the coordinates of joints for each person were normalized and the pose was classified using SVM with the normalized coordinates. Experimental results on video images of CCTV camera on ordinary alleys showed that the proposed method relatively recognized a specific behavior.
KSP 제안 키워드
CCTV Camera, Joint Coordinates, Position information, behavior recognition, everyday life, joint position, perspective effect, recognize people, the body, video images