Reflection-Robust Stereo Visual-Inertial Odometry via Kalman-Filtered Ground-Height Estimation
Organizations: School of Electrical Engineering, Korea Advanced Institute of Science and Technology, Daejeon, 34141, South Korea
Abstract
Visual-inertial odometry relies on visual features that correspond to 3D points with consistent projections across sequential frames. In reflective indoor-floor environments, however, specular floor patterns can be detected and triangulated as virtual features below the physical ground plane, producing unreliable feature associations and inconsistent visual residuals. This paper proposes a lightweight reflection-aware front-end for feature-based stereo VIO. Rather than suppressing all floor-region features, the proposed method estimates the ground height in real time and rejects only feature candidates whose inferred 3D positions are below the estimated ground height. A floor-region mask restricts the validation process to reflection-prone regions, while stereo height uncertainty and pose uncertainty are used to construct frame-level ground-height measurements. A scalar Kalman filter with Normalized Innovation Squared (NIS) gating then provides a stable ground-height estimate, and an uncertainty-aware safety margin is used for selective below-ground virtual feature rejection. Experiments on real reflective-floor indoor sequences show that the proposed front-end generally improves trajectory accuracy in reflection-dominant environments while maintaining accurate ground-height estimates and recovering from actual ground-height changes.
Figures & tables
| Sequence | XY Area | Path Length | Height Change |
|---|---|---|---|
| [m 2 ] | [m] | ||
| E4 | 1499.96 | 184.07 | No |
| E11 | 944.83 | 236.55 | No |
| N3 | 1996.42 | 198.54 | No |
| N7 | 2976.38 | 161.41 | Yes |
| N7 (hand) | 3294.29 | 152.18 | Yes |
| Method | E4 | E11 | N3 | N7 | N7 (hand) | Yonsei 1 | Yonsei 2 | Yonsei 3 |
|---|---|---|---|---|---|---|---|---|
| VINS-Fusion | 0.722/0.405 | 1.238 / 0.364 | 0.845/0.558 | 0.265 / 0.125 | 0.538/0.224 | 1.394/0.726 | 1.416/0.577 | 0.853/ 0.473 |
| VINS-Fusion (reject floor) | 0.592/0.313 | 1.365/0.444 | 0.762/0.366 | 0.510/0.229 | 0.520/0.247 | 1.564/0.915 | 1.313/0.528 | 0.894/0.620 |
| VINS-Fusion + Ours | 0.335 / 0.115 | 1.317/0.424 | 0.460 / 0.215 | 0.338/0.138 | 0.501/0.222 | 1.168/0.660 | 1.303/0.511 | 0.742 /0.503 |
| DynaVINS | – | – | 2.557/0.685 | 2.147/1.109 | – | – | 1.335/0.705 | 0.987/0.405 |
| DynaVINS (reject floor) | 0.693/0.387 | 0.762/0.333 | 0.809/0.393 | 0.497/0.107 | – | 1.232/0.674 | 1.433/0.706 | 0.932/0.463 |
| DynaVINS + Ours | 0.537/0.328 | 0.759/0.289 | 0.550/0.348 | 0.351/0.102 | 0.625/0.250 | 1.168/0.632 | 1.227/0.511 | 0.615/0.340 |
| Sequence | Mean | Std | 30% | 50% | 90% |
|---|---|---|---|---|---|
| [m] | [m] | [m] | [m] | [m] | |
| E4 | 0.029 | 0.044 | 0.009 | 0.018 | 0.066 |
| E11 | 0.088 | 0.127 | 0.023 | 0.064 | 0.207 |
| N3 | 0.050 | 0.049 | 0.018 | 0.032 | 0.098 |
| Yonsei 1 | 0.044 | 0.052 | 0.018 | 0.029 | 0.085 |
| Yonsei 2 | 0.038 | 0.048 | 0.014 | 0.026 | 0.079 |
Visual Odometry
Visual odometry estimates camera or robot motion from successive images, enabling navigation and mapping when external positioning is unavailable.
See all 60 papers