An Algorithm for Micro-localization in Large Public Buildings

Abstract

This paper presents an algorithm for people localization in large public buildings using Bluetooth Low Energy (BLE) beacons, Near-Field Communication (NFC) passive tags and information from specially designed Building Information Model (BIM). The proposed algorithm does not require any pre-data collection. An adaptive Kalman filter is used to decrease the noise in Received Signal-Strength Index (RSSI) raw measurements from beacons. To calculate a fine-grained user's position we find intersection points between rings, which inner and outer radiuses depends on fluctuations in RSSI signals from beacons. Then, for calculated intersection points, we obtain the optimal number of clusters using ANN clustering and inter-clusters entropy. These cluster canters are potential candidates for the position of the visitor. Using dead reckoning, we find a circle-shaped area in which the visitor is expected to be. Only clusters with centres located within this area are taken into account. If several points are in this area, the winner point is one, which belongs to clustering process that gives minimum inter-cluster entropy. The tests show that the localization error is below 1.5 m for all simulated and real world test scenarios.