Date of Award
Master of Science
Department of Electrical and Computer Engineering
Stuart Kurkowski, PhD
The Air Force has increasingly invested in persistent surveillance platforms gathering a large amount of surveillance video. Ordinarily, intelligence analysts watch the video to determine if suspicious activities are occurring. This approach to video analysis can be a very time and manpower intensive process. Instead, this thesis proposes that by using tracks generated from persistent video, we can build a model to detect events for an intelligence analyst. The event that we chose to detect was a suspicious surveillance activity known as a casing event. To test our model we used Global Positioning System (GPS) tracks generated from vehicles driving in an urban area. The results show that over 400 vehicles can be monitored simultaneously in real-time and casing events are detected with high probability (43 of 43 events detected with only 4 false positives). Casing event detections are augmented by determining which buildings are being targeted. In addition, persistent surveillance video is used to construct a social network from vehicle tracks based on the interactions of those tracks. Social networks that are constructed give us further information about the suspicious actors flagged by the casing event detector by telling us who the suspicious actor has interacted with and what buildings they have visited. The end result is a process that automatically generates information from persistent surveillance video providing additional knowledge and understanding to intelligence analysts about terrorist activities.
DTIC Accession Number
Schmitt, Daniel T., "Automated Knowledge Generation with Persistent Surveillance Video" (2009). Theses and Dissertations. 2559.