Estimating performance limits for an intelligent scene monitoring system (ISM) as a perimeter intrusion detection system (PIDS)

Intelligent Scene Monitoring (ISM) is an evolutionary next step beyond basic Video Motion Detection. An ISM system takes video imagery as an input and is expected to alarm only when a specific user-defined event or sequence of events occurs in the scene. It is expected to work in “busy” scenes, alarming only on the target patterns of behaviour, to the exclusion of all other activity. A fundamental question arises out of using video as a PIDS input. What levels of performance (in terms of Probability of Detection P_D and False alarm Rate FAR)? Video (particularly outdoors) is a complex data source and the image processing task for PIDS is non-trivial in the ISM context. We have investigated the performance limits for ISM systems using neural network classifiers in an outdoors “sterile zone” application. This paper defines the sterile zone analysis and image processing tasks, the study methodology and presents some feasibility results (obtained using seal data). It presents an academic, and experimentally justifiable, viewpoint on the tractability of the problem given a predetermined system performance