A Repetitive Observation Strategy for Recognizing a True Anomaly and Estimating its Position

For many surveillance situations, the use of mobile robots has several advantages over static sensors. One of the advantages is that immediate action or investigation can be undertaken when an anomaly is detected. This paper presents a surveillance mobile robot that uses novelty detection. In this project, a neural network is trained to remember the normal measurements of the robot sensors, and highlight any discrepancies from the normal encountered during the inspection period. The contribution of this paper is a new approach to estimating the possible location of the detected anomaly using sensors which provide directional information. The method reduces the number of false alarms by taking into account the location where sensor measurements are made and not just their value. This paper presents details of the method and experimental results demonstrating its application.