A framework for semi-automatically identifying fully occluded objects in 3D models: Towards comprehensive construction design review in virtual reality. (October 2021)

• Record Type: Journal Article
• Authors: Han, Bing; Ma, Jong Won; Leite, Fernanda
• Is Part Of: Advanced engineering informatics. Volume 50(2021)
• Publication Date: 2021-10
• Subjects: Virtual Reality -- Occluded Objects -- Construction Design Review -- Automation -- Visualization; Computer-aided engineering -- Periodicals; Engineering -- Data processing -- Periodicals; 620.00285
• DOI: 10.1016/j.aei.2021.101398
• Languages: English
• ISSNs: 1474-0346

Abstract

Highlights: Point cloud-based algorithms for automatically identifying occluded objects in 3D building information models. Compatible input and output files with current industry practices. Achieved over 90% recall rate and over 75% precision rate in classifying the visibility status of objects. Validated using building information models of an academic building and an industrial facility. Abstract: Virtual Reality (VR)-based construction design review applications have shown potential to enhance user performance in many research projects and experiments. Currently, visualizing occluded objects in VR is a challenge, and this function is indispensable for construction design review and coordination. This paper proposes an occlusion detection framework that semi-automatically identifies occluded objects in 3D construction models. The framework determines the visibility status of an object by converting the object to a point cloud and comparing the point cloud to the virtual laser scanning result of the original model. It exports models that are interoperable with VR development software so that visualization effects can be easily employed to occluded objects. The authors validated the framework using two building information models. The algorithm achieved a recall rate of 90.30% and a precision rate of 75.05% in a gasoline refinery facility model. It reached a higher 98.06% recall rate and a 97.53% precision rate in an academic building model. This paper contributes to the body of knowledge by proposing a semi-automatic occlusion detection framework and validating that point cloud-based algorithms are appropriate for this classification task.