Bootstrapped Learning of Wi-Fi Access Point in Hybrid Positioning System

摘要

In a GNSS/Wi-Fi/Sensors integrated navigation system,Wi-Fi is responsible for providing absolute positionupdates indoors, at anytime. The triangulation method isgenerally preferred in Wi-Fi positioning for coveringlarge-scale ranges. This method is reliable; however, themain implementation challenge is dealing with the lack ofthe geographic information for the local Wi-Fi accesspoints.This paper explores a bootstrapped algorithm that canderive the AP coordinate by itself; then the derivedknowledge can be maintained locally on the mobiledevice for positioning. If no prior Wi-Fi knowledge isavailable, the engine selects GNSS/sensors positioningsolution as known surveying points so that the coordinatesof the AP can be derived in a real-time.Extensive field tests have been performed using Marvell’sleading smartphone platform to verify the algorithms. Thesystem can derive around 80% of the available Wi-Fiaccess points within an uncertainty of 25 meters. As aresult, GNSS/MEMS/Wi-Fi hybrid positioning accuraciesof 10 meters can be obtained for over 95% of time inindoor walking tests. The bootstrapped method of writingthe Wi-Fi AP database alleviates the dependency forapriori infrastructure knowledge, thus it promotes thepractical deployment of a self-contained indoorpositioning product for the consumer market.