Application of 3D point cloud completion for underwater structures

Yi Horng LAI; Senpeng WU; Zixin LI; Tianxin GUO

doi:10.1117/12.3107260

Application of 3D point cloud completion for underwater structures

Yi Horng LAI^*
, Senpeng WU
, Zixin LI
, Tianxin GUO

^*Corresponding author for this work

Research output: Book Chapters | Papers in Conference Proceedings › Conference paper (refereed) › Referred Conference Paper › peer-review

Abstract

The point cloud completion network has been widely used in point cloud measurement projects. However, underwater structure detection often cannot obtain detailed 3D point cloud data in advance, resulting in the inability to train point cloud completion networks. This study proposes a 3D point cloud completion technique for underwater structures. Based on the 2.5D point cloud's geometric characteristics of underwater structures, the RANSAC algorithm is applied to estimate the reference 3D model of underwater structures. Furthermore, train the PCN network with the characteristics of the reference 3D model. Finally, utilize PCN network to generate a comprehensive 3D point clouds of underwater structures. The effectiveness of the proposed method is verified through actual underwater experiments in harbor. The proposed method realizes the application of 3D point cloud completion for underwater structures without a reference ground truth.

Original language	English
Title of host publication	International Conference on Computer Vision and Image Computing, CVIC 2025
Editors	Luis GOMEZ, Zahid AKHTAR
Publisher	SPIE
ISBN (Electronic)	9798902321002
ISBN (Print)	9798902320999
DOIs	https://doi.org/10.1117/12.3107260
Publication status	Published - 13 Feb 2026
Externally published	Yes
Event	International Conference on Computer Vision and Image Computing, CVIC 2025 - Hong Kong, Hong Kong, China Duration: 31 Oct 2025 → 2 Nov 2025

Publication series

Name	Proceedings of SPIE - The International Society for Optical Engineering
Volume	14070
ISSN (Print)	0277-786X
ISSN (Electronic)	1996-756X

Conference

Conference	International Conference on Computer Vision and Image Computing, CVIC 2025
Abbreviated title	CVIC 2025
Country/Territory	Hong Kong, China
City	Hong Kong
Period	31/10/25 → 2/11/25

Bibliographical note

Publisher Copyright:
© 2026 SPIE.

Funding

This paper was supported by Xiamen University Tan Kah Kee College (Grant no. JGH 2025035). The author thanks the Systems Engineering Laboratory, National Sun Yat-Sen University for providing the sonar point cloud data.

Keywords

PCN
point cloud completion
RANSAC
underwater structures

Access to Document

10.1117/12.3107260

Cite this

@inproceedings{eed34aae43fb4a698f38b07054828204,

title = "Application of 3D point cloud completion for underwater structures",

abstract = "The point cloud completion network has been widely used in point cloud measurement projects. However, underwater structure detection often cannot obtain detailed 3D point cloud data in advance, resulting in the inability to train point cloud completion networks. This study proposes a 3D point cloud completion technique for underwater structures. Based on the 2.5D point cloud's geometric characteristics of underwater structures, the RANSAC algorithm is applied to estimate the reference 3D model of underwater structures. Furthermore, train the PCN network with the characteristics of the reference 3D model. Finally, utilize PCN network to generate a comprehensive 3D point clouds of underwater structures. The effectiveness of the proposed method is verified through actual underwater experiments in harbor. The proposed method realizes the application of 3D point cloud completion for underwater structures without a reference ground truth.",

keywords = "PCN, point cloud completion, RANSAC, underwater structures",

author = "LAI, \{Yi Horng\} and Senpeng WU and Zixin LI and Tianxin GUO",

note = "Publisher Copyright: {\textcopyright} 2026 SPIE.; International Conference on Computer Vision and Image Computing, CVIC 2025, CVIC 2025 ; Conference date: 31-10-2025 Through 02-11-2025",

year = "2026",

month = feb,

day = "13",

doi = "10.1117/12.3107260",

language = "English",

isbn = "9798902320999",

series = "Proceedings of SPIE - The International Society for Optical Engineering",

publisher = "SPIE",

editor = "Luis GOMEZ and Zahid AKHTAR",

booktitle = "International Conference on Computer Vision and Image Computing, CVIC 2025",

address = "United States",

}

LAI, YH, WU, S, LI, Z & GUO, T 2026, Application of 3D point cloud completion for underwater structures. in L GOMEZ & Z AKHTAR (eds), International Conference on Computer Vision and Image Computing, CVIC 2025., 140700Z, Proceedings of SPIE - The International Society for Optical Engineering, vol. 14070, SPIE, International Conference on Computer Vision and Image Computing, CVIC 2025, Hong Kong, Hong Kong, China, 31/10/25. https://doi.org/10.1117/12.3107260

Application of 3D point cloud completion for underwater structures. / LAI, Yi Horng; WU, Senpeng; LI, Zixin et al.
International Conference on Computer Vision and Image Computing, CVIC 2025. ed. / Luis GOMEZ; Zahid AKHTAR. SPIE, 2026. 140700Z (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 14070).

Research output: Book Chapters | Papers in Conference Proceedings › Conference paper (refereed) › Referred Conference Paper › peer-review

TY - GEN

T1 - Application of 3D point cloud completion for underwater structures

AU - LAI, Yi Horng

AU - WU, Senpeng

AU - LI, Zixin

AU - GUO, Tianxin

PY - 2026/2/13

Y1 - 2026/2/13

N2 - The point cloud completion network has been widely used in point cloud measurement projects. However, underwater structure detection often cannot obtain detailed 3D point cloud data in advance, resulting in the inability to train point cloud completion networks. This study proposes a 3D point cloud completion technique for underwater structures. Based on the 2.5D point cloud's geometric characteristics of underwater structures, the RANSAC algorithm is applied to estimate the reference 3D model of underwater structures. Furthermore, train the PCN network with the characteristics of the reference 3D model. Finally, utilize PCN network to generate a comprehensive 3D point clouds of underwater structures. The effectiveness of the proposed method is verified through actual underwater experiments in harbor. The proposed method realizes the application of 3D point cloud completion for underwater structures without a reference ground truth.

AB - The point cloud completion network has been widely used in point cloud measurement projects. However, underwater structure detection often cannot obtain detailed 3D point cloud data in advance, resulting in the inability to train point cloud completion networks. This study proposes a 3D point cloud completion technique for underwater structures. Based on the 2.5D point cloud's geometric characteristics of underwater structures, the RANSAC algorithm is applied to estimate the reference 3D model of underwater structures. Furthermore, train the PCN network with the characteristics of the reference 3D model. Finally, utilize PCN network to generate a comprehensive 3D point clouds of underwater structures. The effectiveness of the proposed method is verified through actual underwater experiments in harbor. The proposed method realizes the application of 3D point cloud completion for underwater structures without a reference ground truth.

KW - PCN

KW - point cloud completion

KW - RANSAC

KW - underwater structures

UR - https://www.scopus.com/pages/publications/105032513104

U2 - 10.1117/12.3107260

DO - 10.1117/12.3107260

M3 - Conference paper (refereed)

AN - SCOPUS:105032513104

SN - 9798902320999

T3 - Proceedings of SPIE - The International Society for Optical Engineering

BT - International Conference on Computer Vision and Image Computing, CVIC 2025

A2 - GOMEZ, Luis

A2 - AKHTAR, Zahid

PB - SPIE

T2 - International Conference on Computer Vision and Image Computing, CVIC 2025

Y2 - 31 October 2025 through 2 November 2025

ER -

Application of 3D point cloud completion for underwater structures

Abstract

Publication series

Conference

Bibliographical note

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this