I cannot provide extensive information about projects in progress, because the work involved is incomplete, under review, or in some cases sensitive. For details on aspects of my work, refer to the list of publications in my résumé. I hope the following paragraphs provide a hint of what my current research activities and intersts are.
I am a member of a team that is currently working on developing a massive distributed video surveillance system that can automatically detect activities and events of critical importance to the enterprise of securing public areas and critical infrastructure. The system integrates standard surveillance capabilities such as video viewing, recording, and Pan-Tilt-Zoom camera control with video content analysis for alarm generation, and video processing for real-time tracking as an alternative to manual PTZ camera control.
The completed system will become operational by the end of 2006 at a site with over one hundred cameras. More than ten distinct event detection and activity recognition modules are being prepared. A small-scale installation with tens of cameras is already used for testing at a local site.
Specific areas that I am involved in are: High-performance low-level video processing (video conditioning and feature extraction); Mid-level vision (mostly for feature extraction); Tracking in cluttered scenes; Drivers and controllers for Pan-Tilt-Zoom systems.
A number of lower-level tasks such as vehicle detection and tracking have been addressed by some of my and my colleagues' research to a point that now allows me to attack problems in the realm of data collection for transportation events with categories that are more abstract, for example, collection of statistics on lane changes, U-turns, headway times and other events of interest to traffic engineers.
I am using spectral embedding methods and relaxed (incomplete, semi-metric) distance functions in order to cluster vehicle motion trajectories into groups that reveal the structure of traffic intersections. At this stage, I am particularly interested in the spatial layout of such intersections, but the next step is to develop a richer model that would enable the detection and classification of events that are exemplified by their dynamic rather than spatial structure.
The two major components of this project were real-time vehicle tracking at intersections and short-term vehicle trajectory extrapolation with fast collision testing for the purpose of detecting potential vehicle collsions and issuing warnings (as part of an intelligent transportation infrastructure environment) or starting pre-accident video recording. ARIVL Intersection Monitoring
Driver distraction and inattention are implicated in a large number of traffic accidents. The goal of a driver inattention detection system is to detect lapses in driver concentration by monitoring events such as microsleeps, yawns, or as was the case in this project, talking on a cell phone. AIRVL Driver Fatigue Monitoring