EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Adaptive Point Cloud Downsampling Based on 3-D ROI Detection for Improved Perceptual Quality

Chaeyun Lim, Seunghyun Park, Hyunhee Park

IEEE Access · 2026 · 14 · 68039-68053

DOI: 10.1109/access.2026.3689278 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Chaeyun Lim und 2 weiteren Autor:innen. Die Arbeit nennt 23 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Research Foundation of Korea (NRF) grant funded by Korean Government [Ministry of Science and ICT (MSIT)] and the Ministry of Trade, Industry and Ener…, Hansung University.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This study proposes an ROI-based adaptive downsampling method for efficient transmission of 3D point cloud data. Conventional downsampling techniques that treat all points equally, such as voxel grid sampling (VGS) and farthest point sampling (FPS), do not distinguish important regions in a scene, leading to the loss of visually meaningful information. The proposed method detects 2D region of interest (ROIs) through multi-axis projections, reconstructs them into a 3D ROI, and adaptively allocates the available point budget by preserving a higher point density in ROI regions while more aggressively reducing non-ROI regions. As a result, it maintains visual quality and the stability of downstream tasks even under low-quality transmission conditions. Experimental results on the Microsoft Voxelized Upper Bodies (MVUB) and 8i Voxelized Full Bodies (8iVFB) datasets demonstrate competitive overall perceptual similarity and improved robustness on classification, detection, and pose estimation tasks, particularly at lower quality levels. These results indicate that the proposed method can enhance the quality and reliability of point cloud transmission in bandwidth-constrained environments and suggest that it is an effective and practical alternative for point cloud streaming scenarios that prioritize visually important information.

Förderung

  • National Research Foundation of Korea (NRF) grant funded by Korean Government [Ministry of Science and ICT (MSIT)] and the Ministry of Trade, Industry and Ener…
  • Hansung University

Zitieren

APA

Lim, C., Park, S., & Park, H. (2026). Adaptive Point Cloud Downsampling Based on 3-D ROI Detection for Improved Perceptual Quality. IEEE Access, 14, 68039-68053. https://doi.org/10.1109/access.2026.3689278

Vancouver

Lim C, Park S, Park H. Adaptive Point Cloud Downsampling Based on 3-D ROI Detection for Improved Perceptual Quality. IEEE Access. 2026;14:68039-68053. doi:10.1109/access.2026.3689278

BibTeX
@article{Lim2026,
  author = {Lim, Chaeyun and Park, Seunghyun and Park, Hyunhee},
  title = {Adaptive Point Cloud Downsampling Based on 3-D ROI Detection for Improved Perceptual Quality},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {68039--68053},
  doi = {10.1109/access.2026.3689278}
}
Kostenlos starten →