Factor Graph Optimization for Robust Indoor Positioning: A Data-Driven Approach Integrating Audio Ranging and Pedestrian Dead Reckoning

This work presents a novel approach for indoor positioning by integrating a data-driven audio ranging algorithm with pedestrian dead reckoning (PDR) constrained by magnetic information (MI). The proposed system leverages convolutional neural networks (CNNs) to process time-domain audio signals by transforming them into spectrograms, thus enhancing the accuracy of signal arrival time estimation in complex indoor environments. The PDR system operates at 20 Hz and adapts to various smartphone usage postures by combining sensor data from microphones, BLE, and IMU sensors. In order to improve robustness, the proposed system incorporates multiple robust factors within particle filter (PF) and factor graph optimization (FGO) algorithms, thus effectively mitigating abnormal observations and reducing positioning errors. The experimental results demonstrate that the proposed system achieves high positioning accuracy, with 95% of errors being within 1 m and maximum errors not exceeding 1.7 m across different smartphones, making it a viable solution for precise indoor positioning in real-world scenarios.