Multi-Level Degeneracy Compensation for a Robust LiDAR-Inertial Localization System
作者
Chuqiao Lu,Jie Luo,Linqiu Gui,Shuheng Yu
出处
期刊:2021 International Conference on Electronic Information Engineering and Computer Science (EIECS)日期:2025-09-26卷期号:: 155-159
标识
DOI:10.1109/eiecs67708.2025.11283510
摘要
Degenerate geometry weakens LiDAR constraints and degrades observability in LiDAR-inertial localization. We propose a multi-level compensation framework that couples Normal Distributions Transform (NDT) with an Error-State Kalman Filter (ESKF). At the point level, a point-to-distribution degeneracy factor derived from local Gaussian models weights observation residuals. At the direction level, a global information matrix is spectrally analyzed to detect weakly observable axes, and the Kalman update is projected onto the non-degenerate subspace. Experiments on MulRan dataset and our self-collected dataset show consistent localization-accuracy improvements of 12.8% and 17.1% over the ESKF+NDT baseline.