A collaborative approach to heading estimation for smartphone-based PDR indoor localisation

Abstract

Pedestrian dead reckoning (PDR) is widely used for indoor localisation. Its principle is to recursively update the location of the pedestrian by using step length and step heading. A common method to estimate the heading in PDR is to use magnetometer measurements. However, unlike outdoor environments, the Earth's magnetic field is strongly perturbed inside buildings making the magnetometer measurements unreliable for heading estimation. This paper presents a new method to reduce heading estimation errors when magnetometers are used. The method consists of two components. The first component uses a machine learning algorithm to detect whether a heading estimate is within a specific error margin. Only heading estimates within the error margin are retained and passed to the second component, while the other estimates are discarded. The second component uses data fusion to average the heading estimates from multiple people walking in the same direction. The rationale of this component is based on the observation that magnetic perturbations are often highly localised in space and if multiple people are walking in the same direction, then only some of their magnetometers are likely to be perturbed. Data fusion between users can be carried out in a distributed manner by using a consensus algorithm with information sharing over wireless links. We tested the performance of our method using 92 datasets. The method is shown to provide an average heading estimate error of approximately 2°, which is more than 6-fold lower than the error of the heading estimate based only on raw magnetometer measurements (without any filtering and fusion). Assuming highly accurate step-length observation, the improved heading estimation leads to an average localisation accuracy of 55cm, which is an 80% improvement over PDR localisation using only raw magnetometer measurements.