Deep Learning for Cleaning Cultural Heritage Point Clouds

Hayward, Luc and Marais, Patrick and Wegner, Jan Dirk (2024) Deep Learning for Cleaning Cultural Heritage Point Clouds, Proceedings of 5th Southern African Conference for Artificial Intelligence Research (SACAIR'24), 2-6 December 2024, Bloemfontein, South Africa, 52-67.

Text (PDF paper and proceedings front matter) Pages from SACAIR24_vol_II.pdf - Published Version Download (19MB)

Abstract

Laser scanning technology is often used in the Cultural Heritage domain to capture the 3D structure of a site, with each scan consisting of a set of 3D point coordinates — a point cloud. Before these point clouds can be utilized to build a complete 3D surface model, unwanted points must be removed. While manual point cloud cleaning is time-consuming, previous work has shown promise in automating parts of the process. This study builds on a previous approach which interprets point cloud cleaning as a segmentation task accomplished via binary point classification, applied to individual point clouds. This approach uses a basic Random Forest (RF) classifier with hand-crafted features, is designed to clean scans one by one via incremental per scan training, and requires a complex graph-based post-processing step to achieve acceptable results. By contrast, we leverage modern point-based deep learning models to directly learn useful features, and develop a framework that processes the fully registered set of point clouds, rather than cleaning scans individually. Our method focuses on purely geometric attributes, uses a few-shot fine-tuning approach and, unlike the single scan method, does not require segmentation post-processing to improve results. Under this scheme, users label 2.5 − 50% of an unlabelled scan, and a model is trained to label the rest. We assess three deep learning point-based models (Pointnet++, KPConv, Point Transformer) along with a baseline Random Forest model, focusing on speed, accuracy, and the reduction of total labelling effort. Our findings reveal that modern deep learning requires minimal human labelling, with up to 85% reduction in total labelling effort. KPConv stands out for its efficiency with less human input, while Random Forests work best for simpler scenes. This study highlights deep learning’s effectiveness in reducing manual labour in point cloud cleaning in the cultural heritage domain.

Item Type:	Conference paper
Additional Information:	ISBN : 978-0-7961-6069-0 (e-book)
Uncontrolled Keywords:	Point Clouds, Deep Learning, Cultural Heritage
Subjects:	Computing methodologies > Machine learning Applied computing
Alternate Locations:	https://2024.sacair.org.za/proceedings
Date Deposited:	12 Dec 2024 07:19
Last Modified:	12 Dec 2024 07:19
URI:	https://pubs.cs.uct.ac.za/id/eprint/1711

Actions (login required)

View Item