Novel crowdsourced Fingerprint database update strategy using clustering and pattern matching techniques

Abstract

In fingerprint-based localization system, the difference between the radio map and RSS signatures in the real world is a major factor for deteriorating localization accuracies. Unfortunately, such differences could happen for many reasons including AP relocation, addition, deletion or new walls. Recalibrating the radio map, however, is not only labor-intensive and but also time-consuming. Thus, autonomous fingerprint database update may have a huge impact on building more cost-effective location estimation/tracking system since it could prevent frequent radio map calibration and recalibration. In order for the current RSS signatures to be applied to the radio map automatically, we propose a novel path detection algorithm which uses clustering-based localization and pattern recognition technique. Our approach is capable of finding accurate path even if the path includes areas over which RSS conditions have been changed since our system employs a whole sequence of RSS signatures instead of a single RSS signature. Our extensive sets of experiments report up to 5 times more accurate localization estimations comparing to clustering-based localization techniques even under changed RSS conditions are given.