DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | Choi, Ho-Jin | - |
dc.contributor.advisor | 최호진 | - |
dc.contributor.author | Jin, Sou-Young | - |
dc.contributor.author | 진소영 | - |
dc.date.accessioned | 2013-09-12T01:49:22Z | - |
dc.date.available | 2013-09-12T01:49:22Z | - |
dc.date.issued | 2012 | - |
dc.identifier.uri | http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=509485&flag=dissertation | - |
dc.identifier.uri | http://hdl.handle.net/10203/180465 | - |
dc.description | 학위논문(석사) - 한국과학기술원 : 전산학과, 2012.8, [ v, 31 p. ] | - |
dc.description.abstract | This thesis studies the problem of human action recognition by using Microsoft`s XBox Kinect. Our research is motivated by the observation that an ``hand waving" action is not easily recognized in a ``standing positioned hand waving" action video and a ``walking positioned hand waving" action video when both two videos are used to learn the ``hand waving" action. This is mainly due to the limitation of the current action video representation approaches, which use whole range of a human body as features. We present a novel human action recognition approach that detects characterizing body-parts for an action and concentrates on them to improve the performance of the action recognizer. To detect the essential body-parts, the proposed approach (i) represents an action video with twenty separate body-parts; (ii) finding patterns for each body-part; (iii) clustering body-parts by the pattern lengths; and (iv) decide the vectors belong to the top ranked clusters as essential. Once essential body-parts are detected, the action recognizer concentrates on the body-parts to recognize an action. Our experimental results show that the essential vectors are well identified when three clusters are formed and only the top ranked cluster is used. It is also proven that using the essential body-parts improves the performance of the action recognizer. The main contribution is that the proposed approach is able to detect essential body-parts in an automatic way. This helps an action recognizer learns an action without setting a limitation on training videos. Hence, our proposed approach will ultimately minimize the required human efforts to achieve the goal of vision-based human action recognition research. | eng |
dc.language | eng | - |
dc.publisher | 한국과학기술원 | - |
dc.subject | human action recognition | - |
dc.subject | essential body-parts | - |
dc.subject | longest common subsequence (LCS) algorithm | - |
dc.subject | 사람의 행동인식 | - |
dc.subject | 필수적인 신체부분 | - |
dc.subject | 최대 공통 부분 수열 (LCS) 알고리즘 | - |
dc.subject | K-평균 알고리즘 | - |
dc.subject | K-means algorithm | - |
dc.title | Essential body-part detection for human action recognition | - |
dc.title.alternative | 사람의 행동 인식을 위한 필수적인 신체부분 탐지 | - |
dc.type | Thesis(Master) | - |
dc.identifier.CNRN | 509485/325007 | - |
dc.description.department | 한국과학기술원 : 전산학과, | - |
dc.identifier.uid | 020104435 | - |
dc.contributor.localauthor | Choi, Ho-Jin | - |
dc.contributor.localauthor | 최호진 | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.