PROFusion: Robust and Accurate Dense Reconstruction via Camera Pose Regression and Optimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Dong, Siyan, Wang, Zijun, Cai, Lulu, Ma, Yi, Yang, Yanchao
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908862681972736
author Dong, Siyan
Wang, Zijun
Cai, Lulu
Ma, Yi
Yang, Yanchao
author_facet Dong, Siyan
Wang, Zijun
Cai, Lulu
Ma, Yi
Yang, Yanchao
contents Real-time dense scene reconstruction during unstable camera motions is crucial for robotics, yet current RGB-D SLAM systems fail when cameras experience large viewpoint changes, fast motions, or sudden shaking. Classical optimization-based methods deliver high accuracy but fail with poor initialization during large motions, while learning-based approaches provide robustness but lack sufficient accuracy for dense reconstruction. We address this challenge through a combination of learning-based initialization with optimization-based refinement. Our method employs a camera pose regression network to predict metric-aware relative poses from consecutive RGB-D frames, which serve as reliable starting points for a randomized optimization algorithm that further aligns depth images with the scene geometry. Extensive experiments demonstrate promising results: our approach outperforms the best competitor on challenging benchmarks, while maintaining comparable accuracy on stable motion sequences. The system operates in real-time, showcasing that combining simple and principled techniques can achieve both robustness for unstable motions and accuracy for dense reconstruction. Code released: https://github.com/siyandong/PROFusion.
format Preprint
id arxiv_https___arxiv_org_abs_2509_24236
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle PROFusion: Robust and Accurate Dense Reconstruction via Camera Pose Regression and Optimization
Dong, Siyan
Wang, Zijun
Cai, Lulu
Ma, Yi
Yang, Yanchao
Robotics
Computer Vision and Pattern Recognition
Real-time dense scene reconstruction during unstable camera motions is crucial for robotics, yet current RGB-D SLAM systems fail when cameras experience large viewpoint changes, fast motions, or sudden shaking. Classical optimization-based methods deliver high accuracy but fail with poor initialization during large motions, while learning-based approaches provide robustness but lack sufficient accuracy for dense reconstruction. We address this challenge through a combination of learning-based initialization with optimization-based refinement. Our method employs a camera pose regression network to predict metric-aware relative poses from consecutive RGB-D frames, which serve as reliable starting points for a randomized optimization algorithm that further aligns depth images with the scene geometry. Extensive experiments demonstrate promising results: our approach outperforms the best competitor on challenging benchmarks, while maintaining comparable accuracy on stable motion sequences. The system operates in real-time, showcasing that combining simple and principled techniques can achieve both robustness for unstable motions and accuracy for dense reconstruction. Code released: https://github.com/siyandong/PROFusion.
title PROFusion: Robust and Accurate Dense Reconstruction via Camera Pose Regression and Optimization
topic Robotics
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2509.24236