Two-stage bridge point cloud segmentation by fusing deep learning and heuristic methods

Zhang, Tian, Chen, Haonan, Li, Pengfei and Li, Haijiang ORCID: https://orcid.org/0000-0001-6326-8133 2025. Two-stage bridge point cloud segmentation by fusing deep learning and heuristic methods. Measurement 250 , 117125. 10.1016/j.measurement.2025.117125 Item availability restricted.

PDF - Accepted Post-Print Version Restricted to Repository staff only until 26 February 2026 due to copyright restrictions. Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (1MB)

Abstract

The point cloud acquired from the data acquisition equipment is not segmented by components, and reverse engineering without component segmentation has limited value. Existing heuristic methods achieve high segmentation accuracy but are computationally intensive. On the other hand, deep learning models are quick but often lack accuracy and depend on limited datasets. To address these issues, this paper introduces a fusion method that utilizes results trained on an easily created synthetic dataset to initially segment the point cloud roughly, aiming for a segmentation accuracy and intersection over union ratio of 80 % and 70 %, respectively. Subsequently, a streamlined heuristic method is applied to comprehensively segment the point cloud. The verification results of the instance indicate that this approach achieves the same high level of accuracy (≥99 %) as heuristic methods but increases the speed of segmentation by approximately 2.52 times. The method involves using a synthetic dataset, derived from real point cloud data, in conjunction with the fusion method and selecting a segmentation network that is optimized for simple synthetic datasets.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Schools > Engineering
Publisher:	Elsevier
ISSN:	0263-2241
Date of First Compliant Deposit:	4 April 2025
Date of Acceptance:	24 February 2025
Last Modified:	07 Apr 2025 13:46
URI:	https://orca.cardiff.ac.uk/id/eprint/177403

Actions (repository staff only)

Edit Item