FR-PDR: Factor Graph and Residual Attention Neural Network Improved Pedestrian Dead Reckoning with Shoulder-Mounted IMU

The pedestrian dead reckoning (PDR) plays an increasingly important role in the personnel position-ing of the industrial Internet of Things. However, PDR based on shoulder-mounted inertial measurement units (IMUs) face issues such as the divergence of heading er-rors over time and inaccurate step length estimation. To address these challenges, this paper proposed FR-PDR, which is improved by factor graph and residual attention neural network (RANN) with shoulder-mounted IMU. On the one hand, factor graph (FGO) combined with quasi-static magnetic constraint (QSMC) integrates attitudes based on magnetic field and inertia, effectively correcting the cumulative error of gyroscopes. On the other hand, RANN constructs an accurate step length estimation model under different walking speeds using only sparse training datasets. Through experimental verification, the method proposed has reduced the cumulative heading error to 0.24°/min and the single-step mean absolute error to 0.04m. The collaborative optimization of these two errors ulti-mately reduces the positioning error to 0.81%. The experimental results indicate that the method proposed signifi-cantly improves the positioning accuracy of PDR system with shoulder-mounted IMU, greatly enhances pedestrian indoor navigation performance.