东南大学学报（自然科学版）2011，Vol.41Issue(5)：962-966,5.DOI:10.3969/j.issn.1001-0505.2011.05.013
基于时空兴趣点和主题模型的动作识别
Action recognition based on space-time interest points and topic model
胡斐 1罗立民 2刘佳 1左欣3
作者信息
- 1. 东南大学计算机科学与工程学院,南京,210096
- 2. 武警江西省总队司令部,南昌,330025
- 3. 上海交通大学图像处理与模式识别研究所,上海,200240
- 折叠
摘要
Abstract
A novel human action recognition method based on a probabilistic topic model is proposed. The method uses a bag of words to describe several human common actions including running, jumping, handwaving etc. By extracting local features of space-time interest points. Using the probabilistic topic model can make the action category tag correspond to latent variables, and the inference for latent variables can be used to categorize the entire video actions. In addition, the algorithm can recognize multiple actions by dividing each interest point into different action categories. The method is fast and accurate, and achieves good results in practice. Large numbers of experiments show that the recognition rate of the algorithm is more than 85%.关键词
动作识别/时空兴趣点/主题模型/词袋Key words
action recognition/ space-time interest points/ topic model/ bag of words分类
信息技术与安全科学引用本文复制引用
胡斐,罗立民,刘佳,左欣..基于时空兴趣点和主题模型的动作识别[J].东南大学学报（自然科学版）,2011,41(5):962-966,5.
