원문정보
초록
영어
To realize context aware applications for smart home environments, it is necessary to recognize function or usage of objects as well as categories of them. On conventional research for environment recognition in an indoor environment, most of previous methods are based on shape models. In this paper, we propose a method for recognizing objects focused on the relationship between human actions and functions of objects. Such relationship becomes obvious on human action patterns when he or she handles an object. To estimate object categories by using action patterns, we represent such relationship in Dynamic Bayesian Networks (DBNs). By learning human actions toward objects statistically, objects can be recognized. Finally, we performed experiments and confirmed that objects can be recognized from human actions without shape models.
목차
1. Introduction
2. Overview.
3. Extracting Features of Human Actions.
3.1. Tracking Face and Hands
3.2. Features of Human Action
4. Recognizing Object Using DBNs
4.1. A Typical DBN
4.2. DBNs of Human Actions and Objects
4.3. Observed and State Transition Probability Matrices
4.4. DBNs of Object Regions
5. Experiments
5.1. Experiments without Overlapped Regions
5.2. Experiments with Overlapped Region
5.3. Discussions
6. Conclusion
References