Abstract
Aiming at the problem of insufficient robustness of factor graph models in multi-sensor integrated positioning, especially in GNSS and pedestrian dead reckoning (PDR) fusion scenarios, where existing methods suffer from significant positioning accuracy degradation due to abnormal sensor interference, cumulative error propagation, and dynamic environmental noise, and where current robustness improvement solutions mostly rely on manual experience or high-precision sensors while lacking systematic optimization frameworks, this paper proposes an adaptive fusion scheme based on cross-time-step constraints. By establishing an inertial sensor noise hard-truncation mechanism and a global historical state reverse-correction model, we propose a multi-step kinematic constraint framework based on state correlation, which enhances the consistency between sensor heterogeneous characteristics and the positioning model, suppresses the long-term impact of PDR cumulative errors on positioning, and reduces the interference of abnormal GNSS observations on system states.Simulation and field experiments show that the proposed scheme can effectively fuse low-cost smartphone sensor data in complex environments including urban canyons and signal-occluded areas, achieving highly robust integrated navigation and providing theoretical support for seamless positioning in dynamic interference scenarios.
Funding Information
  • Scientific Research Fund of the Zhejiang Provincial Department of Education (CXJJ2024076)
  • Scientific Research Innovation Fund of Hangzhou Dianzi University (2023QNRC001)
  • Zhejiang Province "14th Five-Year" Second Batch Graduate Provincial Teaching Reform Project (JGCG2024202)

This publication has 33 references indexed in Scilit: