摘要
在时空兴趣点以及3D-SIFT描述的基础上,采用概率主题模型进行动作识别,结果表明:概率主题模型不仅能够实现对视频中单个动作的识别,而且对复杂情况下的视频也有一定识别能力。
A novel method of action recognition based on probabilistic topic model is proposed to recognize several human common actions,on the basis of spatial-temporal interest point and 3D-SIFT description.The result shows that the probabilistic topic model has the recognition not only to single action in video,but also to the sequential video action under complex condition.
出处
《长春工程学院学报(自然科学版)》
2011年第3期113-117,共5页
Journal of Changchun Institute of Technology:Natural Sciences Edition
关键词
人体行为识别
时空兴趣点
主题模型
human action recognition
spatial-temporal interest point
topic models