DECL: A circular inference method for indoor pedestrian localization using phone inertial sensors

Abstract

Pedestrian dead reckoning plays an important role in indoor pedestrian localization applications. Although this approach has a notable advantage that no extra infrastructure is required, it also suffers an issue known as the drift, which means the estimation errors accumulate and ultimately may make the result unreliable. In this paper, we propose a circular inference method applying online learning in order to reduce such drift errors. Map information is used as prior knowledge and identified land marks are used as triggers for learning processing. A multidimensional optimization algorithm is designed and used in learning phase to efficiently tune the estimation parameters. On the basis of the design we implement an end system running on smartphones and use it in the evaluation experiments. The results show that the proposed method can effectively improve the accuracy and reliability of the localization system.