닫기
Loading..

전자정보연구정보센터 ICT 융합 전문연구정보의 집대성

학술대회 프로시딩

홈 홈 > 연구문헌 > 학술대회 프로시딩 > 한국정보과학회 학술대회 > KCC 2020

KCC 2020

Current Result Document : 0 / 0

한글제목(Korean Title) 1D CNN에 기반한 빠른 비디오 행동 인식
영문제목(English Title) Fast video action recognition based on 1D CNN
저자(Author) Jeonghyun Joo   Heeyoul Choi   주정현   최희열                          
원문수록처(Citation) VOL 47 NO. 01 PP. 0880 ~ 0882 (2020. 07)
한글내용
(Korean Abstract)
영문내용
(English Abstract)
Human action recognition from videos is one of the fundamental tasks in the computer vision field. Because videos can be seen as a sequence of still images, we can take the strategy that captures spatial features from each frame and fuses them to reflect temporal information. While spatial features are mostly captured by a well-known powerful image recognizer, convolutional neural networks (CNNs), fusing mechanism can vary by model. One possible way is using recurrent neural networks (RNNs) which is specialized to process sequential data. However, it damages parallelism because it requires to be computed sequentially. In this study, we suggest a video action recognition model that combines spatial features through an additional convolutional layer applied along the temporal dimension. Our model shows similar performance at significantly reduced computation time compared to the model based on RNN.
키워드(Keyword)                              
원문 PDF 다운로드