Title:
PCIM: Deep Learning-Based Point Cloud Information Modeling Framework
PCIM: Deep Learning-Based Point Cloud Information Modeling Framework
Author(s)
Park, Jisoo
Advisor(s)
Cho, Yong Kwon
Editor(s)
Collections
Supplementary to
Permanent Link
Abstract
Although building information modeling (BIM) has been widely used in entire construction projects for data exchange between stakeholders, it sometimes rarely represents the current state of the construction sites because most of the information models are built at the design phase. For this reason, several studies have implemented a process to generate as-built BIM by leveraging laser-scanned point cloud data, referred to as Scan-to-BIM. However, the conventional Scan-to-BIM process is usually performed manually or semi-manually, which is time-consuming and labor-intensive. Moreover, captured reality and measurements such as actual dimensions, shapes, colors, and damages of the target subjects that the original point cloud retains can be disregarded during the solid modeling process in Scan-to-BIM. To address this issue, this research proposes an object-oriented information modeling framework for point cloud data, named point cloud information modeling (PCIM).
The main objective of this dissertation is to develop artificial intelligence (AI)-driven information modeling framework for point cloud data. To this end, this study 1) presents methods of classifying construction objects and their properties and 2) proposes a data schema to represent the classified information with an object-based hierarchy structure. At this time, the scope of this research is limited to building construction area, but the data schema can be extended to other jobs such as mechanical, electrical, and plumbing (MEP) engineering. The findings of this research may rebound to the benefit of stakeholders of construction projects considering that point clouds play an essential role in the construction management phase. Since this research will provide an automated information modeling solution for point cloud data, stakeholders that apply the proposed approach will save time to generate an as-is construction site model. Moreover, this research may fill the gaps in current studies on object classification in 3D by leveraging extended input channels such as laser intensity and material index. As this paper presents the concept of PCIM, various follow-up studies are expected to be additionally derived.
Sponsor
Date Issued
2020-12-02
Extent
Resource Type
Text
Resource Subtype
Dissertation