It's the Human that Matters: Accurate User Orientation Estimation for Mobile Computing Applications

TL;DR

Humaine is a system that accurately estimates user orientation relative to Earth using inertial sensors, working reliably indoors and outdoors without prior setup, significantly outperforming existing methods.

Contribution

The paper introduces Humaine, a novel system that estimates user orientation accurately without prior configuration, applicable in diverse indoor and outdoor scenarios with arbitrary phone positions.

Findings

Achieves median accuracy of 15° in diverse scenarios

Outperforms state-of-the-art by 558% in accuracy

Works without user intervention or specific phone placement

Abstract

Ubiquity of Internet-connected and sensor-equipped portable devices sparked a new set of mobile computing applications that leverage the proliferating sensing capabilities of smart-phones. For many of these applications, accurate estimation of the user heading, as compared to the phone heading, is of paramount importance. This is of special importance for many crowd-sensing applications, where the phone can be carried in arbitrary positions and orientations relative to the user body. Current state-of-the-art focus mainly on estimating the phone orientation, require the phone to be placed in a particular position, require user intervention, and/or do not work accurately indoors; which limits their ubiquitous usability in different applications. In this paper we present Humaine, a novel system to reliably and accurately estimate the user orientation relative to the Earth coordinate system. Humaine requires no prior-configuration nor user intervention and works accurately indoors and outdoors for arbitrary cell phone positions and orientations relative to the user body. The system applies statistical analysis techniques to the inertial sensors widely available on today's cell phones to estimate both the phone and user orientation. Implementation of the system on different Android devices with 170 experiments performed at different indoor and outdoor testbeds shows that Humaine significantly outperforms the state-of-the-art in diverse scenarios, achieving a median accuracy of $15^\circ$ averaged over a wide variety of phone positions. This is $558\%$ better than the-state-of-the-art. The accuracy is bounded by the error in the inertial sensors readings and can be enhanced with more accurate sensors and sensor fusion.