EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

V3Reg: Model Integrating Visual Information for Extreme Low Overlap Point Cloud Registration

Yaxiong Li, Yifan Hou, Qisong Yang, Dongdong Guan

Remote Sensing · 21.06.2026 · 18 (12) · 2050

DOI: 10.3390/rs18122050 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Remote Sensing, online erschienen am 21.06.2026. Verfasst von Yaxiong Li und 3 weiteren Autor:innen. Die Arbeit nennt 42 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Shaanxi Provincial Natural Science Foundation, Foundational and Frontier Innovation Program of Rocket Force University of Engineering.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Extremely low overlap leads to severely scarce local geometric correspondences across frame pairs. Pure geometric descriptors—encoding merely low-level shape signatures—inherently fail to impose sufficient constraints for reliable transformation estimation when matches become critically sparse, rendering registration fundamentally fragile. While recent red-green-blue-depth (RGB-D) attempts have explored visual augmentation, they predominantly rely on low-level chromatic statistics or shallow convolutional neural network (CNN) features, underutilizing the rich hierarchical semantics inherent in RGB imagery. We present V3Reg, a robust registration framework that pioneers the integration of large-scale vision foundation models (DINOv3) with adaptive cross-modal fusion. Specifically, we extract mid-to-deep semantic features (Layer 11) from DINOv3 to transcend low-level texture limitations, and propose a Task-Aware Channel-Wise Gated Adaptive Fusion (TACGAF) module that dynamically calibrates geometric-visual contributions via registration-error-guided channel-wise gating. To rigorously evaluate ultra-low-overlap robustness, we reconstruct RGBD-ZeroMatch, a benchmark with controllable overlap ratios ranging from 1% to 20%. Extensive experiments demonstrate that V3Reg achieves 99.6% Feature Matching Recall and 96.3% Registration Recall on standard benchmarks. Notably, it maintains 50.2% Registration Recall at merely 5% overlap, outperforming prior methods by over 18 percentage points.

Förderung

  • National Natural Science Foundation of China
  • Shaanxi Provincial Natural Science Foundation
  • Foundational and Frontier Innovation Program of Rocket Force University of Engineering

Zitieren

APA

Li, Y., Hou, Y., Yang, Q., & Guan, D. (2026). V3Reg: Model Integrating Visual Information for Extreme Low Overlap Point Cloud Registration. Remote Sensing, 18(12), 2050. https://doi.org/10.3390/rs18122050

Vancouver

Li Y, Hou Y, Yang Q, Guan D. V3Reg: Model Integrating Visual Information for Extreme Low Overlap Point Cloud Registration. Remote Sensing. 2026;18(12):2050. doi:10.3390/rs18122050

BibTeX
@article{Li2026,
  author = {Li, Yaxiong and Hou, Yifan and Yang, Qisong and Guan, Dongdong},
  title = {V3Reg: Model Integrating Visual Information for Extreme Low Overlap Point Cloud Registration},
  journal = {Remote Sensing},
  year = {2026},
  volume = {18},
  number = {12},
  pages = {2050},
  doi = {10.3390/rs18122050}
}
Kostenlos starten →