Improving the Dependability of Distributed Surveillance Systems Using Diverse Redundant Detectors

Sensor networks nowadays employed in critical monitoring and surveillance applications represent a relevant case of complex information infrastructures whose dependability needs to be carefully assessed. Detection models based on Event Trees provide a simple and effective mean to correlate events in Physical Security Information Management (PSIM) systems. However, as a deterministic modeling approach, Event Trees are not able to address uncertainties in practical applications, like: 1) imperfect threat modelling; 2) sensor false alarms. Regarding point (1), it is quite obvious that real-world threat scenarios can be very variable and it is nearly impossible to consider all the possible combinations of events characterizing a threat. Point (2) addresses the possibility of missed detections due to sensor faults and the positive/nuisance false alarms that any real sensor can generate. In this chapter we describe two techniques that can be adopted to deal with those uncertainties. The first technique is based on Event Tree heuristic distance metrics. It allows to generate warnings whenever a threat scenario is detected and it is similar to the ones in the knowledge base repository. The second technique allows to measure in real-time the estimated trustworthiness of event detection based on: a) sensors false alarm rates; b) uncertainties indices associated to correlation operators. We apply those techniques to case-studies of physical security for metro railways.