A Cooperative Machine Learning Approach for Pedestrian Navigation in Indoor IoT

Area of Science:

• Mobile Computing
• Sensor Networks
• Machine Learning

Background:

• Accurate localization of mobile users is crucial for various applications.
• Pedestrian dead reckoning (PDR) offers infrastructure-free indoor localization using device sensors.
• Heading estimation in PDR is challenging indoors due to magnetometer perturbations.

Purpose of the Study:

• To develop a novel cooperative system for enhanced PDR-based localization.
• To improve heading accuracy by addressing magnetometer measurement perturbations.
• To leverage machine learning and consensus algorithms for cooperative data fusion.

Main Methods:

• A machine learning approach detects and filters perturbed magnetometer measurements.
• A consensus algorithm aggregates users walking in the same direction.
• Distributed data fusion combines measurements from aggregated users for accurate heading estimation.

Main Results:

• The proposed system significantly reduces heading errors in indoor environments.
• Cooperative data fusion enhances localization accuracy compared to individual PDR.
• The system demonstrates robustness against sensor failures through pre-filtering.

Conclusions:

• The integration of machine learning and consensus algorithms offers a novel solution for cooperative PDR.
• This infrastructure-free, distributed approach improves indoor localization accuracy and reliability.
• The method is the first to combine ML with consensus for cooperative PDR, outperforming existing techniques.