GeoNLF: Geometry guided Pose-Free Neural LiDAR Fields
This addresses the challenge of pose-free neural reconstruction for LiDAR data in autonomous driving and robotics, offering a more robust solution compared to existing methods, though it builds incrementally on prior NeRF and geometric techniques.
The paper tackles the problem of synthesizing LiDAR point clouds without relying on precomputed poses, which are often inaccurate, by proposing GeoNLF, a hybrid framework that alternates between neural reconstruction and geometric pose optimization. The result shows superiority in novel view synthesis and multi-view registration on datasets like NuScenes and KITTI-360, with concrete improvements in handling sparse-view inputs and large-scale point clouds.
Although recent efforts have extended Neural Radiance Fields (NeRF) into LiDAR point cloud synthesis, the majority of existing works exhibit a strong dependence on precomputed poses. However, point cloud registration methods struggle to achieve precise global pose estimation, whereas previous pose-free NeRFs overlook geometric consistency in global reconstruction. In light of this, we explore the geometric insights of point clouds, which provide explicit registration priors for reconstruction. Based on this, we propose Geometry guided Neural LiDAR Fields(GeoNLF), a hybrid framework performing alternately global neural reconstruction and pure geometric pose optimization. Furthermore, NeRFs tend to overfit individual frames and easily get stuck in local minima under sparse-view inputs. To tackle this issue, we develop a selective-reweighting strategy and introduce geometric constraints for robust optimization. Extensive experiments on NuScenes and KITTI-360 datasets demonstrate the superiority of GeoNLF in both novel view synthesis and multi-view registration of low-frequency large-scale point clouds.