I-Corps: 3D Scanning Tool for Reconstruction Via Uncertainty Quantification

University Of Texas At Austin, Austin TX

Investigators

Abstract

The broader impact/commercial potential of this I-Corps project is the development of an interactive 3D scanning tool that targets novice users. The goal of the proposed technology is to bridge 3D reconstruction via 3D scanners, which typically require extensive expertise, to novice users with no background in geometry processing. 3D scanners generate depth scans that are partial surfaces of the underlying object. A 3D reconstruction typically requires registering depth scans under different camera poses, known as the registration procedure, and reconstructing a complete surface from the registered scans, known as surface reconstruction. The performance of registration heavily depends on the structure of how scans overlap with each other. The proposed technology uses a principled uncertainty quantification framework that may accurately predict the uncertainties of 3D reconstruction from depth scans. It was determined that the predicted uncertainties provide an effective means to guide scan planning that minimizes the number of scans and improves reconstruction quality. This technology may reduce the 3D acquisition time and improve the reconstruction quality. Applications include reconstructions of human teeth for dentists to perform surgery planning, reconstructions of buildings, bridges, and other architectural artifacts for damage inspection, 3D reconstructions of traffic accidents for crime analysis, and 3D reconstructions of mechanical parts for model checking. This I-Corps project is based on the development of uncertainty quantification algorithms for multi-scan registration and reconstruction and view planning algorithms based on quantified uncertainties. The proposed technology is designed for quantifying the uncertainties of the current 3D reconstruction, automating reconstruction completeness checking, online scan planning to reduce the uncertainties of reconstruction, enhancing completeness, and minimizing scanning time. The goal is to integrate the proposed technology into existing software packages for 3D reconstruction, making them easy to use for novice users with no background in geometry processing. The proposed technology addresses the limitations of existing scan-based 3D reconstruction techniques, especially the long acquisition period and the lack of effective reconstruction quality assessment methods, problems that have been well-documented in the research literature. Research findings examining the proposed technology suggest that efficient uncertainty quantification techniques may enhance the quality of 3D reconstruction and reduce scanning time. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.