Agent-based security system for user verification

Abstract

We present a security system consisting of an arbitrary number of entries, sensors and agents. The intelligent integrated system is based on user modeling, i.e. models of their previous entries, using several levels of description and several single machine learning agents. The system is specialized for high-security tasks, e.g. controlling entry to a headquarter and for scenarios including terrorist attacks or kidnapping, when the attackers formally bypass the existing sensors, but fail to imitate normal movement of an employee. An experimental setup was implemented to verify the successfulness of the approach. © 2009 IEEE.