Computer-Aided Physical Security

CS Dept. Head Candidate

Terrance Boult
Lehigh University

Tuesday, March 27, 2003
11 a.m. - 12 noon
Fuller Labs 320

Abstract

Video and other sensors have long been used as a part of physical security. Physical security has been largely immune to the IT revolution -- the data, if it is looked at at all, is usually sent to video monitors and LED displays where a security guard is supposed to "notice" significant events in those flickering displays. This talk will discuss the general area of computer aided physical security, why it has been slow to have been adopted, our research results over the past 7 years advancing this area, and some of the ongoing challenges in securing important national infrastructure.

A plenitude of computer and video systems aspects need to be addressed to support physical security; the talk will briefly review the necessary concepts and components. As the key sensor data to be processed is video, we will look more closely at issues related to two important vision subtasks: detection/tracking and human identification. This will be a general overview of the research and does not presume a background in security or video systems.

The talk will summarize our work on video surveillance, starting with the DARPA VSAM (video surveillance and monitoring) project through our current work on detection and tracking. We examine the problem constraints, explaining the strong demand they place on the vision subsystem. The presentation will include examples of "difficult" situations and review the techniques that have allowed our system to track within these settings and will provide examples of tracking snipers.

We will briefly overview the detection/tracking system design and give examples of the key steps. We will explain one key component, QCC (quasi- connected components) which combines gap filling, multi-level thresholding-with- hysteresis, region definition and shape morphology. The talk then briefly discusses a generalized approach to ROC analysis for surveillance problems that can significantly reduce the "costs" of gathering the necessary ground-truth and performance data while still supporting parameter optimization.

A topic of even greater media attention today is biometrics, a.k.a. human identification or HID. We discuss general human identification issues and review a few basic biometrics suitable for use at a distance. We then describe our ongoing efforts in novel sensor for HID, and in evaluation of face-based HID systems under varying sensor and weather conditions. This will include a brief discussion of our HID evaluation paradigm and some of the pit-falls one needs to avoid in such evaluations. With a better understanding of their limitations, we discuss the role HID systems can play in current and future computer-aided physical security systems.

The end of the talk will return to the high level systems view discussing the issues of systems/network architecture, watermarking, information privacy and encryption.

Host

Computer Science Dept Head Search Committee

Refreshments will be served.

Maintained by webmaster@wpi.edu
Last modified: Sep 27, 2006, 16:05 EDT