Consistency-Guided Fusion of Asymmetric Quantitative and Qualitative Sensor Information for Urban 3D Localization in Vehicular IoT Systems
TL;DR
Results demonstrate that consistency-triggered qualitative constraints can complement metric sensor fusion without directly introducing uncertain visual measurements, providing a practical mechanism for improving the robustness and vertical stability of heterogeneous localization systems.
Abstract
High-precision urban 3D localization is a critical foundation for vehicular Internet of Things (IoT) applications, yet conventional Global Navigation Satellite System (GNSS)-based localization is vulnerable to signal obstruction, multipath effects, and non-line-of-sight propagation in complex environments. To improve localization reliability, this paper proposes a consistency-guided quantitative–qualitative fusion (CG–QQF) framework with vision-based terrain constraints. The framework integrates heterogeneous quantitative sensors, including an absolute positioning source, inertial measurement unit (IMU), wheel encoders, and a steering angle sensor, for continuous metric state estimation, while a monocular camera provides qualitative terrain-slope information. Rather than treating visual perception as a direct metric observation, the proposed method introduces it as a conditional structural constraint that is activated only when it is consistent with the quantitative estimate, thereby regularizing the localization solution and suppressing vertical drift. Although ultra-wideband (UWB) positioning is adopted as the absolute positioning source in the experimental platform, it serves as a generic positioning module and can be replaced by GNSS-based techniques such as real-time kinematic (RTK) and precise point positioning (PPP). In an indoor scaled proof-of-concept experiment over a controlled four-lap dataset, CG–QQF achieves a 3D RMSE of 0.0575 m and a vertical MAE of 0.0042 m. Its 3D RMSE is approximately 4.0% lower than quantitative sensor fusion (QSF), 37.9% lower than absolute-positioning/inertial fusion (ABS–INS), and 49.9% lower than vision-assisted quantitative fusion (VA–QF). These results demonstrate that consistency-triggered qualitative constraints can complement metric sensor fusion without directly introducing uncertain visual measurements, providing a practical mechanism for improving the robustness and vertical stability of heterogeneous localization systems.