Covert Human Behaviour Detection
Punarjay Chakravarty, Dave Rawlinson, Ray A. Jarvis · 2007
This paper describes a method by which suspicious, covert human movements can be distinguished from ordinary activity. A network of xed cameras observes an arbitrary indoor environment. The positions of moving, visible entities are separately tracked in a global planar map using multiple particle lters. Some objects can be visually identied as friendlies (guards). The motion of other objects is classied as covert (suspicious) or overt (normal) using two metrics that measure overall visibility of the chosen route and visibility with respect to guards’ position. In other words, the system measures to what extent a person is trying to hide from the guards. This paper is the latest in a series of related works that together describe how a network of xed cameras can learn to identify intruders and guide mobile robots to intercept.