RKHS-BA: A Robust Correspondence-Free Multi-View Registration Framework with Semantic Point Clouds

TL;DR

This paper introduces RKHS-BA, a robust multi-view registration framework that leverages continuous landmark representations in RKHS, enabling correspondence-free pose estimation with high robustness and generalization across noisy and semantic-rich data.

Contribution

It presents a novel RKHS-based multi-frame Bundle Adjustment framework that operates without explicit correspondences, improving robustness and generalization in multi-view registration tasks.

Findings

Achieves highly robust pose estimation in noisy scenes

Demonstrates strong generalization with various semantic inputs

Effective in multi-view registration, odometry, and mapping

Abstract

This work reports a novel multi-frame Bundle Adjustment (BA) framework called RKHS-BA. It uses continuous landmark representations that encode RGB-D/LiDAR and semantic observations in a Reproducing Kernel Hilbert Space (RKHS). With a correspondence-free pose graph formulation, the proposed system constructs a loss function that achieves more generalized convergence than classical point-wise convergence. We demonstrate its applications in multi-view point cloud registration, sliding-window odometry, and global LiDAR mapping on simulated and real data. It shows highly robust pose estimations in extremely noisy scenes and exhibits strong generalization with various types of semantic inputs. The open source implementation is released in https://github.com/UMich-CURLY/RKHS_BA.