【摘要】 人体行为识别是计算机视觉的研究难点和热点,主流的研究框架包括行为特征提取、人体行为表示和识别算法3个方面,目前简单场景下的人体简单动作的识别已基本得到解决,而复杂场景下的行为识别仍面临很多困难。对近几年人体行为识别的发展做了比较详细的研究,从人体行为识别的研究范畴、特征提取以及行为模型等方面综述了目前复杂场景下人体行为识别的研究方法。与已有的相关综述文献不同的是,文中结合了近三年国内外人体行为识别领域中新的研究热点和成果,如姿态特征的提取和表示、基于稀疏编码和卷积神经网络的人体行为表示方法等。最后阐述了该领域目前存在的困难以及可能的发展趋向。更多还原

【Abstract】 Human action recognition has become a hot and difficult spot currently in the domain of computer vision.The framework of mainstream methods includes visual feature detection,action representation and action classification.Action recognition in simple scenes has been implemented at present.This paper introduced in detail the research of human action recognition in realistic scenes from perspectives of research scope,feature detection,and action modeling.Unlike several recent published researches,we analyzed the state-of-the-arts and advances of this field,such as pose estimation,sparse coding based or deep learning based human action representation etc.Finally,the problems,difficulties as well as possible solutions were discussed.更多还原