Research at the Computer Vision Laboratory at the University ofMaryland has focussed on developing algorithms and systems that can lookat humans and recognize their activities in near real-time. Our earlierimplementation while quite successful, was restricted to applicationswith a fixed camera. In this paper we present some recent work thatremoves this restriction. Such systems are required for machine visionfrom moving platforms such as robots, intelligent vehicles, andunattended large field of regard cameras with a small field of view. Ourapproach is based on the use of a deformable shape model for humanscoupled with a novel variant of the condensation algorithm that usesquasi-random sampling for efficiency. This allows the use of simplemotion models which results in algorithm robustness, enabling us tohandle unknown camera/human motion with unrestricted camera viewingangles. We present the details of our human tracking algorithms and someexamples from pedestrian tracking and automated surveillance
展开▼
