GrassMA: Graph-Based Semi-Supervised Manifold Alignment for Indoor WLAN Localization

Abstract

Wireless local area network (WLAN) fingerprinting has been extensively studied for indoor localization due to the formidable deployability of WLAN in indoor environment, but one major bottleneck of its practical implementation is the extensive calibration effort required to construct the radio map through the fingerprinting which is time consuming and labor intensive. In response to this compelling problem, we newly design the graph-based semi-supervised manifold alignment approach which relies on the concept of graph construction to construct a cost-efficient radio map with a small number of labeled fingerprints. In addition, the unlabeled user traces are also considered to compensate for the sparsity of the raw radio map as well as enhance its robustness to the environmental change. The extensive experiments conducted in an actual indoor WLAN environment demonstrate the performance improvement (by 27.48% at most with respect to the confidence probability of errors within 3 m) by our system compared with the existing ones using the fingerprints solely.