Click here for full text:
Argus: Visual Sensing for Large-Scale Tracking
Low, Colin
HPL-2005-12R1
Keyword(s): remote sensing; computer vision; tracking; scene analysis
Abstract: An architecture and theory are provided for tracking many thousands of people with a large scale network of cameras. Cameras are not used to identify individuals; instead they generate a robust and characteristic visual signature, and probabilistic techniques derived from air and naval theatre tracking are used to construct maximum likelihood tracks. A novel feature of the system is that the topology of the camera network is deduced from very noisy data, and the system performs deductively without an explicit a priori representation of the camera network. Detailed simulation confirms the theoretical analysis and shows the algorithms can perform well. Applications are likely to include reconstructive analysis of criminal acts in public places, route analysis in commercial zones, and security in transport termini.
22 Pages
Back to Index
| 
Printable version
