WiFi fingerprint positioning based on clustering in mobile crowdsourcing system

Abstract

WiFi Fingerprint Positioning (WFP) in outdoor scenario needs mass location information including WiFi signal map and GPS (Global Positioning System) information. Generally pre-measured solution can provide high quality data but it needs lots of labor and time. Different from pre-measured solution, crowdsourcing is an economic and efficient way to obtain location information. WFP based on Clustering (WFP-C) in mobile crowdsourcing system is proposed to improve positioning accuracy and reduce computation complexity. WFP-C includes three phases: offline database building, dataset preprocessing and online positioning. In offline database building phase, Android-based APP is developed to collect crowdsourcing data. In dataset preprocessing phase, according to some clustering algorithm, the geography area is divided into several fingerprint clusters which are identified by Position Feature Vectors (PFVs). In online positioning phase, two-stage matching method is proposed. Firstly, the WiFi signal vector is used to match some cluster according to PFVs. And then, the accurate position is calculate using the WiFi signal vector of the cluster. The Android-based APP is installed in smart phones which are carried by ten volunteers. The collected data is used to evaluated our proposal. The experiment compares WFP-C, grid-based WFP and non-cluster WFP. The evaluation results indicate that WFP-C can achieve higher positioning accuracy and low computation complexity.