Surveillance Video Analysis Systems

The analysis of surveillance video data can exploit the inherent peculiarities of the context to advantage, making it possible, in the near future, to automate video surveillance to a considerable extent. At present, the available technology cannot afford to eliminate or even diminish the role of the human component in the crucial task of interpretation. Automating the analysis will require, to begin with, highly robust front end (pre-processing and segmentation) components that function reliably under highly varying illumination and noise conditions, background activity, foreground object occlusions, etc. Next, one requires reliable tracking through occlusions, automated learning of agent appearances and actions, followed by agent and activity classification in real time. We present some of our contributions, in particular, multi-scale background modelling, occlusion resistant tracking and automated discovery of appearance and activities. Our goal is to develop eventually, domain specific real-time VSA systems than can considerably minimize the demands on human participation.

Biography

Completed B.E. from BMS engineering College in Electronics in 1987. MTech from the Electrical Engineering Department of IIT Kanpur over 1987-1989, with a dissertation in the area of Rational Choice Theory, titled "Additive Conjoint Measurement Over Incomplete Orders" that provides a theoretical framework for rational decision making under conditions of incomplete choice information. Was a PhD student at IIT Kanpur's Electrical Engineering Dept over 1989-95, working on a geometrical-topological reformulation of Signal and System Theory, titled "Signals and Systems on Sets" that sought to relax many of the mathematical assumptions made in conventional theory. Worked in IIT Guwahati from 1995 until mid 1999 as assistant professor in the Electronics and Communication Engineering Dept. Moved to IIT Kanpur's Electrical Engineering Department in 1999 and have since concerned myself with problems in the applied areas of image/video processing, computer vision with applications in video surveillance, robot navigation, visual metrology and human computer interfaces, and the mathematical modeling of drug delivery and dissemination in the context of hormone based approaches to male contraception.

University of Ulster

ISRC Seminars