AIs Role in Semantic Segmentation for Data-Driven 3D Models of Heritage Structures

Title

AIs Role in Semantic Segmentation for Data-Driven 3D Models of Heritage Structures

Creator

Battina, Subhadha; Jaganathan, Siva

Description

Using point cloud data from laser scanning and photogrammetry to create three-dimensional models with scan-to-BIM processes has become increasingly common in heritage conservation. During the processing of point clouds, semantically segmenting data can translate captured spatial information into intelligent data structures, enabling classified, accurate, data-driven digital models of heritage structures. Subsequently, digital models are utilized for analytical tasks like structural tests, energy optimization, etc. Artificial Intelligence (AI) has become a promising solution for automating Three-Dimensional Point Cloud Semantic Segmentation (3DPCSS), enabling a faster and more accurate composition of parametric objects within 3D modeling and management systems. However, implementing 3DPCSS solely with AI presents various technical and theoretical challenges. The geometrical complexities inherent in historical structures often necessitate manual segmentation processes or oversimplified representations that miss the unique characteristics of the building. Therefore, selecting an appropriate AI framework for 3DPCSS is essential to ensure accurate results. Multiple factors determine algorithms selection, making it challenging to categorize universal solutions. The paper highlights the key factors: 1) Data collecting tools and technologies, 2) Types of the dataset, 3) Complexity of geometrical elements, and 4) Computational tasks. AI frameworks are typically selected based on the suitability and significance of these factors relative to the projects intent. Very few studies talk about the choices of algorithms. This papers significant contribution is recognizing trends in effective data acquisition strategies through a case study in India. Additionally, it identifies state-of-the-art AI models from the past decade based on a systematic literature study. The paper infers the extensive use and advancement of hybrid approaches tailored to multi-modal data types and the specific needs of heritage projects. The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

Source

Lecture Notes in Civil Engineering;Volume;591 LNCE;pp.793-808

Date

01-01-2025

Publisher

Springer Science and Business Media Deutschland GmbH

Subject

3D Point Cloud Semantic Segmentation (3DPSS); AI in cultural heritage; Data-driven 3D Modeling; Digital Heritage; Scan-to-BIM

Coverage

Battina S., Christ University, Bengaluru, India; Jaganathan S., Christ University, Bengaluru, India

Rights

Restricted Access; Hardcopy may be available in the library

Relation

ISSN: 23662557; ISBN: 978-981964050-8;

Format

online

Language

English

Type

Conference paper

Identifier

https://doi.org/10.1007/978-981-96-4051-5_77

https://www.scopus.com/pages/publications/105003633889?origin=resultslist