CCTV monitoring systems are manually monitored, with typical CCTV control rooms having many screens monitored visually by a team of operators typically working around the clock.
Events can often be missed due to operator fatigue and sheer volume of visual information. BT research is studying how video analytics technology can be deployed as part of an information network to provide automated monitoring and decisions to dramatically improve the efficiency and reduce response time to major events.
A key development so far is pioneering new software that leaves the era of IP-CCTV infrastructure behind and moves towards intelligent IP-based video monitoring and surveillance. The new solution enables incident/events detection and intelligence extraction from a distributed network of cameras in real-time.
Whilst the potential applications are endless, it is particularly useful for analysing crowd movement, behaviour information and accurate counting of moving objects in mixed traffic. For example, when dealing with large crowds, the real-time analytics software can provide a continuous measurement of congestion within a defined area, such as a town centre or railway platform.
Railway platform deployment
The analytics software can be deployed on almost any camera with a view of the platform, irrespective of whether the chosen view results in occlusion between individuals. The presence of trains can be accurately detected, and the system can compensate for perspective distortions and variable lighting conditions induced by train headlights or traffic signal lights, for example. The software has been proven in a live trial in underground platforms during which it operated continuously for over 6 months.
Town centre/Open space deployment
The analytics software has an advanced counting application that can classify moving vehicles, pedestrians and bicycles and provide a real-time count of multiple flows within a pre-defined area. Several unique approaches have been perfected (such as multi-tripwire and semi-appearance modelling) to enable accurate counting from sub-optimal camera positions, as might be expected where analytics are required to use images from legacy CCTV equipment.
The particular advantages lie in the ability to deal with streams of loose pedestrian clusters and slow moving and waiting vehicle queues. The current testbed only requires a standard PC for real-time counting of three inward and three outward traffic flows with a very low error rate. It is also robust to an external environment, including changes in lighting levels, shadows, reflections and moving background objects such as trees and has been proven in a live trial at two busy industrial site entrances throughout four seasons.
