Gradient-Based Fingerprinting for Indoor Localization and Tracking

Abstract

Of the different branches of indoor localization research, WiFi fingerprinting has drawn significant attention over the past decade. These localization systems function by comparing WiFi received signal strength indicator (RSSI) and a pre-established location-specific fingerprint map. However, due to the time-variant wireless signal strength, the RSSI fingerprint map needs to be calibrated periodically, incurring high labor and time costs. In addition, biased RSSI measurements across devices along with transmission power control techniques of WiFi routers further undermine the fidelity of existing fingerprint-based localization systems. To remedy these problems, we propose GradIent FingerprinTing (GIFT) which leverages a more stable RSSI gradient. GIFT first builds a gradient-based fingerprint map (Gmap) by comparing absolute RSSI values at nearby positions, and then runs an online extended particle filter (EPF) to localize the user/device. By incorporating Gmap, GIFT is more adaptive to the time-variant RSSI in indoor environments, thus effectively reducing the overhead of fingerprint map calibration. We implemented GIFT on Android smartphones and tablets, and conducted extensive experiments in a five-story campus building. GIFT is shown to achieve an 80 percentile accuracy of 5.6 m with dynamic WiFi signals.