Grid-Based Bayesian Filtering Methods for Pedestrian Dead Reckoning Indoor Positioning Using Smartphones

Area of Science:

• Computer Science
• Robotics
• Signal Processing

Background:

• Indoor positioning systems commonly use Pedestrian Dead Reckoning (PDR).
• PDR accuracy is degraded by sensor noise, step estimation errors, and direction detection inaccuracies.
• Bayesian filtering and map models are used to mitigate errors in PDR.

Purpose of the Study:

• To explore grid-based Bayesian filters as alternatives to Kalman and particle filters for indoor positioning.
• To introduce and analyze advanced grid-based filters: centroid grid filter and advanced point-mass filter.
• To evaluate the performance of these filters on real-world indoor positioning data.

Main Methods:

• Elaboration of previous grid-based filter work with convolution mask calculations.
• Introduction and implementation of the centroid grid filter and the advanced point-mass filter.
• Comparative analysis of filter configurations using raw sensor data from diverse experimental datasets.

Main Results:

• The proposed grid-based filters demonstrate effectiveness in suppressing errors in indoor positioning.
• Evaluation on datasets from Slovakia, France (2018 IPIN), and Italy (2019 IPIN) shows competitive performance.
• The advanced centroid grid filter and point-mass filter are identified as viable alternatives to particle filters.

Conclusions:

• Grid-based Bayesian filters, particularly the advanced implementations, offer a qualified alternative to particle filters for PDR-based indoor positioning.
• The study provides insights into the advantages, drawbacks, and optimal configurations of these advanced filters.
• The findings contribute to the development of more robust and accurate smartphone indoor positioning technologies.