In this research, we have made progress on the four major areas, namely
(1) The modeling of human actions;
(2) Tracking of human actions;
(3) Analysis of human actions;
(4) Network architecture for distributed human action learning modules.
This work has potential applications in different areas of our everyday life. Firstly, it can be applied in intelligent surveillance systems for security purposes in places where the behaviors of the people are expected to be closely monitored for the detection of any abnormality. Secondly, this research will lead to system interfaces that enable computers and robots to learn the individual and collective behaviors of humans, as well as the interactive behaviors between humans and other objects. Thirdly, the contentbased video database system developed in this research has the potential to help the media industry to handle their
vast amount of human-motion-related materials more effectively, particularly in applications of data indexing, retrieval and storage. Apart from the above applications, the developed technology can also be beneficial to other professional areas such as sport analysis, dance choreography and sign translation.
Our research in relation to human action understanding through learning involves the detailed investigation in the following challenging technical problems:
(1) How to detect the presence of humans in a scene;
(2) How to track the path of the humans in the scene and how to track the motion of the body parts of humans;
(3) How to abstract human actions from video using computational models and how to abstract the behavior information of multiple persons into high level collective behavioral representation;
(4) How to analyze the interaction between humans and objects in the scene;
(5) How to recognize the activity types of humans and detect the abnormal behaviors and how to recognize the identity of the humans in the scene;
(6) How to develop a context-aware video database system that supports the storage, indexing and retrieval of
information based on human-behaviorrelated contents.
Instead of solving these problems using heuristic rules, an approach of learning from demonstration examples is adopted. The complexities of many human actions make them very difficult to be handled analytically or heuristically. The solutions to these technical problems will create the core modules whose flexible combination will form the basis of systems that can fit into different application areas.
| 
