Steps Towards Automatic Activity Classification

Justin Muncaster

Computer Science Department

UC Santa Barbara

Date: Friday, May 9, 2008
Place: Engineering Sciences Building 1001
Time: 2:00 pm — 3:00 pm


Abstract:

The demand for new tools for surveillance and security coupled with the availability of cheap sensors has sparked a large amount of research in automated detection and tracking in recent years. With recent progress in automatic detection and more then four decades of research in target tracking, practical and deployable algorithms for tracking objects in a scene are coming closer to fruition. In light this progress we are seeing that future applications in areas such as surveillance and Human Computer Interaction will require a system to not only accurately track objects in a scene, but also to classify behavior of individuals and groups.

In this talk we give an overview of ongoing work to develop a system capable of classifying activity in a scene. We discuss a practical architecture for such a system and show some qualitative tracking results. We'll go over previous work classifying activity and finally conclude with current work being done to identify humans in a scene. paint, cut and deform 3D mesh models.

JUSTIN MUNCASTER is a PhD candidate in the Computer Science Department at UCSB. His research advisor is Prof. Tobias Hollerer. He was funded by the digital multimedia IGERT grant from Fall 2004 through Spring 2006.