A Novel Fingerprinting Method of WiFi Indoor Positioning Based on Weibull Signal Model

摘要

A number of indoor positioning systems based on WiFi fingerprinting were reported thanks to advantages of this method, such as low cost and extensive availability. The Bayesian fingerprinting method needs learn the radio map of probability distribution of WiFi signal strengths over the space of interest through a training phase. Traditionally, the histogram method was used for calculating probability distribution, and it required an adequate number of WiFi samples, which caused a long time taken in the training phase. This study first analyzes the temporal variation of WiFi received signal strength indication (RSSI) at a specific location, and proposes the Weibull signal model for representing the probability density of temporal variation of WiFi RSSI observables. Then, in the positioning phase, the Weibull-based probability density is utilized for Bayesian estimation to resolve the positioning solution. This method is proposed to reduce the required number of RSSI samples for learning probability distribution, and hence improve the effi?ciency of fingerprinting database training. This method is implemented on Android commodity smartphone, and is evaluated in office building environments. Experiment results show that this method reduces the work loading of fingerprinting training due to less samples required, and the positioning accuracy is enhanced by 21-35% up to different building environments, compared to the histogram based method even in which more samples are used.