CAJ | 학술논문

人体行为识别是计算机视觉的研究难点与热点，目前大部分研究者主要针对视频中的行为展开研究．然而，人类的视觉往往根据单张图片就可判断图片中发生的行为．基于单张静态图像的人体行为识别，挑战性更大，是近年来人体行为识别研究的一个趋势，更是探索人类视觉奥秘的一个很好切入点．本文对单张静态图像的人体行为识别方法进行梳理，将其分为三类，最后对其未来研究方向进行展望．
인체행위식별시계산궤시각적연구난점여열점，목전대부분연구자주요침대시빈중적행위전개연구．연이，인류적시각왕왕근거단장도편취가판단도편중발생적행위．기우단장정태도상적인체행위식별，도전성경대，시근년래인체행위식별연구적일개추세，경시탐색인류시각오비적일개흔호절입점．본문대단장정태도상적인체행위식별방법진행소리，장기분위삼류，최후대기미래연구방향진행전망．
Human action recognition is a difficult and active research area in computer vision. At present, most of researchers in this field focus on recognizing action from video. However, human can understand human action based on a single picture. Recognize action from Single still images has more challenge and is a trend in action recognition in recent years, but also a good entry point to explore the mysteries of human vision. In this paper, we sort out the methods of recognizing action from single still images and classify these methods into three categories. At last, the future research directions are discussed.