Target segmentation and event detection at video-rate: the EAGLE project

In this paper we present the EAGLE system: new modular, two-layer object detection and tracking system for static camera situations. The first, low-level layer provides feature extraction and tracking capabilities. The second, high-level layer exploits this information using techniques borrowed from artificial intelligence; its flexible architecture is easily adaptable to different surveillance situations and allows for event detection raising (user defined) warnings and alarms. EAGLE uses intelligent scene observation in order to evaluate interaction with user-defined landmarks instead of probabilistic trajectory estimation. The often unpredictable behavior of humans in motion had us abandon that feature. EAGLE needs no detection areas and has the ability to do "warm starts", i.e., start tracking with targets already present in the scene. With this paper we intend to participate in the CREDS competition.