Data mining-based localisation of spatial low-resolution sensors in commercial buildings

We present an approach to infer relative position and orientation of ubiquitous building-installed low-resolution spatial sensors using real-life measurements. Commercial buildings accumulate a "zoo" of heterogeneous devices, as new generations of sensors and emerging modalities become available. The increasing number of building sensors pose challenging commissioning and maintenance tasks that require sensor arrangement information. In this work we focus on sensors which can track objects in their field of view, i.e., spatial sensors. Furthermore, our proposed approach is fined tuned for low-resolution spatial sensors, i.e., where objects can be detected but not identified. We exploit user walking trajectories within building spaces and the resulting sensor signatures to identify spatial relations between sensors. We start by tracking objects through each sensor's field of view, then create link rules between pairs of sensors, and finally, based on the mined link rules, infer relative sensor arrangement. We evaluated our mining and sorting approach using thermopile array sensors in four real-life building scenarios: corridor, T-crossing, meeting room, and a foyer, involving different sensor arrangements and network sizes. Based on data acquired over multiple weeks in each scenario, we found that sensor arrangements could be inferred with accuracies ranging from 77% to 100% on the best days, depending on scenario complexity and use. Our approach can be beneficial to derive and maintain heterogeneous spatial sensor networks, independent of the sensor's communication protocol.