Comparison of FGO and KF for PDR-GNSS Fusion on Smartphone for Diverse Pedestrian Movements

Smartphone-based positioning has gained significant popularity due to its widespread availability. However, its reliance on GNSS as the primary positioning component poses limitations in challenging environments such as urban canyons, tunnels, and indoor areas, where GNSS signals are distorted or missed due to multi-path interference and non-line-of-sight (NLOS). To address this, sensor fusion techniques have been employed, with the integration of MEMS IMU being a common approach due to its complementary features with GNSS. For pedestrian localization, the fusion of PDR and GNSS is a classic architecture that commonly involves the use of the Kalman Filter (KF). Recently, researchers have started to consider the Factor Graph Optimization (FGO) algorithm for the fusion, due to its better accuracy. However, it is still necessary to asses its performance compared to the KF under different situations, such as various pedestrian movements. In this study, experimental evaluations during walking and running activities demonstrate that FGO consistently outperforms KF, resulting in a significant reduction of mean horizontal errors by 30.6% and 10% in walking and running, respectively. These findings highlight that between walking and running modes, the accuracy of FGO is influenced by the magnitude of PDR errors and the level of correlation in GNSS measurements, resulting in varying performance. Overall, FGO demonstrated superior efficacy compared to KF in enhancing the accuracy and reliability of smartphone-based positioning systems, especially in challenging pedestrian scenarios.