Device-clustering algorithm in crowdsourcing-based localization

Device heterogeneity significantly degrades the localization performance of fingerprinting-based localization, especially in the crowdsourcing-based positioning system. Although manual calibration can reduce positional error, the adjustment overhead is extremely heavy and to maintain ever-increasing device types is overly laborious. In this paper, we propose a novel device-clustering algorithm to operate the positioning system based on macro device-cluster (DC) rather than natural device. In this way, the system maintains less device types and the localization accuracy is improved obviously. The experimental result of different combination indicates the optimal operating flow is to combine DC and kernel density estimator when the tracking device is known and add the linear transformation phase when device is unknown.