Robust event classification for a fiber optic perimeter intrusion detection system using level crossing features and artificial neural networks

Abstract

Discriminating between intrusion and nuisance events without compromising sensitivity is a key performance parameter for any outdoor perimeter intrusion detection system. This is especially the case for intrusion and nuisance events which may have a similar impact on a perimeter fence. In this paper, a robust event classification system using features based on level crossings is presented for the detection and recognition of intrusion and non-intrusion events in an outdoor fence-mounted intrusion detection system for a range of operating environments and fence styles. The proposed classification system is applied to a distributed fiber-optic Mach Zehnder (MZ) mounted on a perimeter fence. It consists of a pre-processing stage employing high resolution time-frequency distribution, a novel event detection and feature extraction scheme based on level crossings, and a classification algorithm using a supervised neural network. Experimental results are presented showing accurate classification of different intrusion and non-intrusion events such as fence-climbing, fence-cutting, stone-throwing and stick-dragging. These results demonstrate the robustness of the proposed algorithm for various types of fence fabric and operating environments.