Mapping normal sensor measurement using regions

Novelty detection, which is highlighting deviation from the normal condition, is a very useful capability for a surveillance mobile robot. In order to perform novelty detection, a robot must first learn what are normal sensor measurements in the environment. Then, during surveillance the robot can recognized any unusual measurement by comparing to what it has learned before. However, a sensor measurement which is unusual at one location could be normal at another. This paper presents a method by which a normal sensor measurement is associated with the area or region where it is observed. The method reduces the storage required for data from different types of sensors. It also reduces the number of false alarms when performing a surveillance task. This paper provides details of the novelty detection scheme and gives experimental results demonstrating its application.