Turning Surveillance Video into Structured Information and Actionable Events

Network-centric smart video surveillance has witnessed a surge of interest from both diverse business sectors and research communities over the last decade. BT´s recent transformation from a dominant fixed line Telco in the UK to a leading networked ICT service provider worldwide has spurred our researches in this and allied field so as to provide value-added services and solutions, addressing the increasing needs of security, safety, cost reduction and intelligence gathering from multi-national corporate customers. Our research has been of both an applied nature, creating end-to-end robust solutions here and now, and as an enabler to track state-of-the-art technology trend allowing for innovative business propositions for emerging applications. In this presentation, I´ll discuss several recent research projects within our Visual Computing and Multimedia Understanding Theme. These include real-time crowd congestion gauging; abnormal activities visualisation and detection in nonlinear subspace; skimming video from news genre to home videos; mixed traffic counting at distance, etc. In each case, a brief review is given in respect of the project´s objective, practical challenges, technical approach and evaluation procedure. Video examples will then be shown to demonstrate the promising results obtained in realistic video scenarios, and to point out any opening problems. Where appropriate we will also discuss the issues how to bridge the gap between a prototype video analytic function and its applicability in real-time real-world surveillance environment.