Semi-Supervised Learning for Indoor Hybrid Fingerprint Database Calibration With Low Effort

Abstract

The interest of indoor localization based on the IEEE 802.11 wireless local area network signal increases remarkably to support pervasive computing applications, but the process of fingerprints calibration, which is point-by-point conducted manually, is time consuming and labor intensive. To address this problem, we propose to use a novel improved semi-supervised manifold alignment approach by integrating the execution characteristic function to reduce both the number of reference points (RPs) and sampling time involved in the radio map construction. Specifically, the radio map is constructed from a small number of calibrated fingerprints and a batch of user traces, which are sporadically collected in the target environment. The user traces enable to compensate for the effort of reducing the calibration cost as well as improving the effectiveness of radio map. In addition, the cubic spline interpolation approach is applied to enrich the radio map with the limited number of RPs. Extensive experiments show that the proposed approach is capable of not only reducing the effort of fingerprints calibration remarkably, but also guaranteeing the high localization accuracy.