Application of Non-homogenous Hmm on Detecting Security Fence Breaching

In this paper, we present a reliable and accurate low computational complexity classifier for a fence intrusion detection sensor. The sensor, a standalone perimeter security sensor, classifies six different types of intrusion attempts on chain-link fences. The designed classifier, a dynamic sequence analyzer followed by a static classifier, obtains more than 95 percent accuracy in intrusion classification. Furthermore, its computational simplicity allows for the classifier to be run on the sensor’s 16 MIPS processor. The sequence analyzer, a HMM with time-dependent transition probability matrix; localizes the most probable time window of different intrusion classes. Then, the static classifier utilizes the analyzer output in intrusion classification. The implemented HMM, a specific set of non-homogenous HMM, has been defined and trained using the Baum-Welch technique. The nonhomogenous HMM accuracy in intrusion localization surpasses homogenous HMM; while the designed classifier achieves three percent improvement comparing to a HMM classifier.