Probabilistic Infrastructureless Positioning in the Pocket

Abstract

With the increasing popularity of smart phones, knowing the accurate position of users has become critical to many context-aware applications. In this paper, we introduce a novel Probabilistic Infrastructureless Navigation (ProbIN) system for GPS-challenging environments. ProbIN uses inertial and magnetic sensors in mobile phones to derive users’ current location. Instead of relying on basic laws of physics (e.g. double integral of acceleration equals to displacement) ProbIN uses a statistical model for estimating the position of users. This statistical model is built based on the user’s data by applying machine learning techniques from the statistical machine translation field. Thus, ProbIN can capture the user’s specific walking patterns and is, therefore, more robust against noisy sensor readings. In the evaluation of our approach we focused on the most common daily scenarios. We conducted experiments with a user walking and carrying the phone in different settings such as in the hand or in the pocket. The results of the experiments show that even though the mobile phone was not mounted to the user’s body, ProbIN outperforms the state-of-the-art dead reckoning approaches.