Validating Passive Localization Methods for Occupancy Sensing Systems in Wireless Environments: A Case Study

In this paper we present an experimental validation of different methods used in a passive indoor localization system for inferring occupancy information on different zones of interest. The experiments were conducted in a lecture building of 6,000 squared meters with 20 classrooms for 6 months. In addition, the teaching computers were used as monitors in order to capture 802.11 traffic. More than 200,000 unique MAC addresses from heterogeneous devices with diverse hardware and software configurations were detected, generating signals with different RSS and temporal patterns. Taking into account the particularities of this experimental environment, this paper first presents a characterization of the passive monitoring system and then analyses different localization methods. Those methods use representations of the RSS measurements based on order vectors, which support the device heterogeneity without requiring special calibration, and distance metrics that are applied to a simple machine learning-based classifier in order to provide satisfying results for occupancy purposes. Our experiments analyse the influence of different time windows and δ values (a threshold used to establish order relationships) on the classification accuracy obtained by the different methods.