HILLS: Hierarchical Indoor Localization for Large-Scale Architectural Complex

So far, small-scale scenarios have always been considered in WiFi fingerprint based indoor localization, where the experiment setting ranges from room-level to floor-level spaces in a single building. In this paper, the goal is to make use of data mining approach to investigate large-scale wireless indoor localization, where an architectural complex is involved for experiments. When small-scale scenarios are extended to large-scale scenarios, the localization accuracy with conventional methods drops significantly because much more intra-interference will be produced. To guarantee a good accuracy in large-scale wireless localization, a novel scheme called Hierarchical Indoor Localization for Large-Scale scenarios (HILLS) is proposed in this paper, where the hierarchical tree is generated based on the information of real architectural complex. Based on the generated tree, a hierarchical classification scheme is proposed for coarse localization based on small-size feature vectors to reduce computational complexity. Furthermore, fine localization follows to achieve a better accuracy by adopting full-size feature vectors and similarity-based classifier. In virtue of inequality of location information in the received signal strength (RSS) from different WiFi access points (APs) and the RSS redundancy at multiple APs, a double weighted similarity-based neighbor (DWSN) is proposed as the classifier in this paper. Experimental results show the proposed scheme achieves an excellent accuracy in large-scale wireless localization.