Indoor Localization using Stable Set of Wireless Access Points Subject to Varying Granularity Levels

Abstract

WiFi-based Indoor Positioning System (IPS) has become a popular approach for providing Location Based Services (LBS) to the smartphone users in indoor environments due to the availability of massive existing WiFi network infrastructure. However, WiFi signal strengths not only vary with time, different smartphone configurations, different ambient conditions including open/closed room, presence/absence of persons and other interfering devices etc. but also it depends on the granularity of positioning. It is easier to achieve good room level accuracy but more precise positioning is difficult. Most of the existing works fix one granularity level for a given context for their work. But the variation of both positioning granularity in terms of how much area is covered by one coordinate location (that is, one grid), and context heterogeneity should be considered as these variations typically affect localization accuracy. Thus, it is necessary to select stable WiFi Access Points (APs) that provide significant localization accuracy for various context and different granularity (grid sizes). Consequently, in this paper, an algorithm that selects stable APs for the most appropriate positioning granularity that minimizes the localization error. Hence, WiFi signal strengths are collected from an entire floor of a building on our University campus. This experimental testbed is divided based on the different size of grids and data has been collected from every possible grid subject to the temporal, ambiance and device heterogeneity for two months. Based on the localization error and ground truth, grids of $1 \times 1$ sq.m. are considered. Using 20 stable APs out of 43 APs, an ensemble method achieved 96.62% localization accuracy and BayesNet provides 4.54% localization error.