Grid-based indoor localization using smartphones

We investigate properties and the accuracy of grid-based Bayesian filters for indoor localization. We use only the compass and a (hardware or software) step detector which are available in commodity smartphones. In the case when a hardware step detector is not present, we propose a reliable step detection algorithm using the accelerometer. Our main goal is to evaluate how the accuracy of a grid-based Bayesian filter is influenced by the discretization of environment, the structure of a grid, the accuracy of the expected step length, and the way a building is modeled. As the result of evaluation, we propose two correction algorithms. The first algorithm is designed to eliminate the influence of discretization. Since the accuracy of the expected step length influences the accuracy of estimated positions computed by a grid-based Bayesian filter, we propose an algorithm that analyzes compass data and changes of estimated positions in order to generate hints to increase or decrease the current value of the expected step length. Evaluation results confirm an improvement when the proposed algorithms are applied.