Model-based analysis of ‘K out of M' correlation techniques for diverse redundant detectors

Sensors are widespread in applications ranging from environmental monitoring to distributed surveillance for physical security. Novel protocols and appropriate topologies enable large networks of cheap smart-sensors with the main objective of providing pervasiveness and resilience. In this paper we provide a model-based analysis of a ‘K out of M’ (‘KooM’) voting approach which can be used to correlate data coming from heterogeneous event detecting devices. The approach is based on the assumption of diverse redundancy on sensor technologies. The Bayesian Network formalism is employed to perform the analysis. The results show that by choosing appropriate correlation logics an optimal trade-off can be achieved among probability of detection, false alarm rate, availability and robustness against spoofing attempts, depending on the specific application. Furthermore, it will be shown that majority voting on detector outputs allows for a high cost effectiveness in obtaining performance improvements.