Performance Analysis of FGO Windowing Strategies for PDR+GNSS Fusion Architecture

Smartphone-based pedestrian navigation has become increasingly popular due to the widespread availability of low-cost Android devices. However, these smartphones often suffer from low positional accuracy caused by inexpensive GNSS antennas and receivers. To address these limitations, fusion mechanisms are employed, with Factor Graph Optimization (FGO) emerging as a prominent method in navigation solutions, particularly for INS+GNSS integration. While FGO has also gained popularity in Simultaneous Localization and Mapping (SLAM) applications, its reliance on batch data processing, referred to as Batch Factor Graph Optimization (BFGO) and which is used as the baseline for this study, results in significant computational overhead, making it impractical for real-time or resource-constrained scenarios. To address these challenges, two windowing-based FGO approaches are proposed: the Naive Windowing FGO (NW-FGO) and the Marginalized Windowing FGO (MW-FGO). By incorporating marginalization, MWFGO retains critical information while reducing computational costs. Both methods are evaluated under two error conditions GNSS multipath and PDR errors to assess their impact on positional accuracy and efficiency. Our findings reveal that optimal window sizes, such as 50s for GNSS errors and 30s for PDR errors, achieve accuracy comparable to BFGO while significantly reducing computational time from 2500 seconds to 38 seconds with MWFGO. Furthermore, MWFGO outperforms NW-FGO, improving mean positional accuracy by 2.18% across varying window sizes, underscoring the effectiveness of marginalization. This study demonstrates the feasibility of windowing-based FGO, particularly MWFGO, in balancing computational efficiency and accuracy. By optimizing window sizes and leveraging marginalization, we provide a robust alternative to FGO, making real-time pedestrian navigation viable in resource-constrained environments.