IPNL Intelligent Positioning & Navigation Lab Search EN 简 繁

Enhancing Smartphone-based Pedestrian Positioning: Using Factor Graph Optimization with Indoor/Outdoor Detection for 3DMA GNSS/Visual-Inertial State Estimation

Ho, H. Y., Ng, H. F., Wen, W., Gu, Y., Hsu, L. T.

Journal of Data Science and Intelligent Systems (2024)

journal Q? Featured page
DOI PDF Code
LIO-SAM ground-truth point clouds, smartphone field imagery, and trajectory comparison for indoor/outdoor pedestrian positioning.
Figures 5, 6, and 14 visual montage: LIO-SAM ground truth, field image matching, and trajectory comparison.

Summary

This paper fuses smartphone 3DMA GNSS and visual-inertial estimates with an indoor/outdoor switching factor in factor graph optimization.

Highlights

  • The system uses SVM-based indoor/outdoor detection to decide how GNSS and VINS measurements contribute to the graph.
  • Hong Kong pedestrian experiments compare smartphone trajectories against LIO-SAM ground truth under indoor/outdoor transitions.

Figures

The SVM classifier turns GNSS features into an indoor/outdoor label that later controls the switching factor.

SVM flowchart for indoor/outdoor classification using training data, validation data, GNSS features, and test data.
Figure 3: SVM flowchart for indoor/outdoor classification.

The factor graph shows where VINS, 3DMA GNSS, and the indoor/outdoor switching factor meet in the optimization problem.

Factor graph connecting VINS factors, 3DMA GNSS factors, switching factors, and state nodes.
Figure 4: FGO structure for loosely coupling GNSS and VINS with the indoor/outdoor switching factor.

The experiment routes define the indoor/outdoor transition scenarios that are later used to compare algorithms.

Google Earth views of experiment trajectories A, B, and C.
Figure 5: Google Earth views of trajectories A, B, and C.

The LIO-SAM ground truth anchors the trajectory comparisons and error plots that follow.

LIO-SAM ground-truth maps for trajectories A, B, and C.
Figure 6: ground truth generated by LIO-SAM for trajectories A, B, and C.

The B2 error plot highlights the transition period where the indoor/outdoor switch reduces the error growth seen in methods without the switch.

B2 absolute positioning error plot with 3DMA GNSS standard deviation and indoor/outdoor indicator.
Figure 13: B2 absolute positioning error comparison with estimated 3DMA GNSS standard deviation and IO indicator.

The companion trajectory plot shows the same effect spatially: the switching version follows the ground truth more closely during the transition.

B2 trajectory comparison against ground truth for VINSMONO, 3DMA GNSS, 3DMA GNSS/VINS FGO, and 3DMA GNSS/VINS-IO FGO.
Figure 14: B2 trajectory comparison.