LR-SGS: Robust LiDAR-Reflectance-Guided Salient Gaussian Splatting for Self-Driving Scene Reconstruction
Ziyu Chen, Fan Zhu, Hui Zhu, Deyi Kong, Xinkai Kuang, Yujia Zhang, Chunmao Jiang
AI summary
Problem
Existing 3D Gaussian Splatting methods for self-driving scenes struggle with complex lighting, high ego-motion, and weak-texture regions because they underutilize LiDAR's geometric and reflectance information, leading to texture inconsistencies and unstable optimization.
Approach
The method introduces a LiDAR-reflectance-guided salient Gaussian representation initialized from geometric and reflectance feature points, coupled with a lighting-invariant reflectance channel and cross-modal gradient alignment to enforce boundary consistency during joint optimization.
Key results
- Superior PSNR and SSIM on Waymo Open Dataset across challenging scenes
- 1.18 dB PSNR gain over OmniRe in complex lighting conditions
- Reduced Gaussian count and faster training time
- Realistic scene editing and scalable synthetic data generation
Why it matters
Provides a robust, efficient reconstruction framework that enhances the reliability and scalability of synthetic data generation for testing and training end-to-end autonomous driving models.
Abstract
Recent 3D Gaussian Splatting (3DGS) methods have demonstrated the feasibility of self-driving scene recon- struction and novel view synthesis. However, most existing methods either rely solely on cameras or use LiDAR only for Gaussian initialization or depth supervision, while the rich scene information contained in point clouds, such as reflectance, and the complementarity between LiDAR and RGB have not been fully exploited, leading to degradation in challenging self- driving scenes, such as those with high ego-motion and complex lighting. To address these issues, we propose a robust and efficient LiDAR-reflectance-guided Salient Gaussian Splatting method (LR-SGS) for self-driving scenes, which introduces a structure-aware Salient Gaussian representation, initialized from geometric and reflectance feature points extracted from LiDAR and refined through a salient transform and improved density control to capture edge and planar structures. Further- more, we calibrate LiDAR intensity into reflectance and attach it to each Gaussian as a lighting-invariant material channel, jointly aligned with RGB to enforce boundary consistency. Ex- tensive experiments on the Waymo Open Dataset demonstrate that LR-SGS achieves superior reconstruction performance with fewer Gaussians and shorter training time. In particular, on Complex Lighting scenes, our method surpasses OmniRe by 1.18 dB PSNR.