目前,基于视觉的人体的行为识别是一个非常活跃的研究领域。它在智能监控、感知接口和基于内容的视频检索等领域具有广泛的应用前景,然而,一些困难仍然减慢了行为识别的发展,比如现实场景中动作往往是从任意角度拍摄。因此与视点无关的行为识别就十分重要,大量的研究者开始致力于行为识别的视点无关性。对视点无关的姿态与运动识别进行了综述。从基于时空特征的方法,基于状态空间的方法,基于降维的方法和基于运动轨迹的方法4个方面分析了研究进展情况,并列举了视点无关行为识别的公共数据集,评价了目前的研究情况,并对未来的研究提出了展望。

Vision-based human action recognition is currently one of the most active research fields. It has many promising applications such as intelligent surveillance, perceptual interface and content-based video retrieval. Even though, some hurdles still slower the development of action recognition, which the actions are often observed from arbitrary camera viewpoints in realistic scene. So view-invariance is important for action recognition, growing number of research groups to pay more attentions to research related to the view-invariant issue. This paper provides a survey on view-invariant recognition of poses and actions. The improvements of this topic in the last several years are discussed in detail from four aspects: space-time based methods, state-space approaches, dimension reduction based methods, and trajectories based methods. Public available standard datasets are presented, and the concluding discussion assesses the progress so far, and outlines some future directions.