Output details
11 - Computer Science and Informatics
Queen Mary University of London
Incremental activity modeling in multiple disjoint cameras.
<23>This paper presents a novel approach based on Time Delayed Probabilistic Graphical Models to understanding video activities from their partial observations captured by multiple cameras with unknown topology, and for the first time, addresses the problem of large scale crowd activity modelling in wide public spaces. Final definitive account and comprehensive formulation of preliminary works including oral presentations at CVPR'09 (10.1109/CVPR.2009.5206827) and ICCV'09 (10.1109/ICCV.2009.5459156) with acceptance rates approx. 4%; outputs from EPSRC/MOD JGS grant EP/E028594/1(JGS1097) and EU FP7 grant SEC-217899; invited IEEE CVPR'12 Tutorial on Security and Surveillance. Key results were reported in the NewScientist (http://www.newscientist.com/article/mg20427385.800-smart-cctv-learns-to-spot-suspicious-types.html).