Intensity-Augmented LiDAR-Visual-Inertial Odometry and Meshing
Yunfeng Hua, Qinyu Liu, Zhongwei Lin, Xiao Gong, Bintao Zhao, Jian Zhang, Tengfei Jiang, Shengjun Shi
Abstract
This paper presents a tightly-coupled LiDAR-Visual-Inertial Odometry (LIVO) system that integrates both LIO and VIO subsystems. The system jointly estimates the state by fusing LiDAR or visual data with Inertial Measurement Units (IMUs). It employs point-to-mesh tracking to optimize LiDAR poses and leverages intensity information from LiDAR point clouds to refine camera pose estimation. The optimized camera pose, derived from VIO, plays a crucial role in texture mapping and 3D geometry synthesis (3DGS) rendering. Our experiments demonstrate a significant improvement in average Peak Signal-to-Noise Ratio (PSNR) compared to existing methods, including R3LIVE, SR-LIVO, and FAST-LIVO. Furthermore, the system features a real-time mapping module implemented on the GPU, utilizing Truncated Signed Distance Function (TSDF) fields for global map maintenance and the Marching Cubes algorithm for mesh extraction. This approach ensures rapid and precise tracking and reconstruction capabilities. Additionally, our system supports real-time remeshing of the global map upon detecting loop closures, thereby enhancing the robustness and accuracy of the overall SLAM process.
BibTeX
@inproceedings{iros2025_intensityaugment,
title = {Intensity-Augmented LiDAR-Visual-Inertial Odometry and Meshing},
author = {Yunfeng Hua and Qinyu Liu and Zhongwei Lin and Xiao Gong and Bintao Zhao and Jian Zhang and Tengfei Jiang and Shengjun Shi and Weiwei Xu},
booktitle = {IROS 2025},
year = {2025}
}