High-accuracy three-dimensional shape acquisition of a large-scale object from multiple uncalibrated camera views

Area of Science:

• Computer Vision
• 3D Reconstruction
• Photogrammetry

Background:

• Accurate 3D shape acquisition from multiple uncalibrated camera views is challenging due to the large number of images required.
• Accumulative errors in 3D reconstruction can arise from processing numerous images sequentially.

Purpose of the Study:

• To deduce error propagation rules in traditional dual-view 3D reconstruction.
• To propose a novel method for controlling accumulative errors in large-scale 3D shape acquisition.
• To enhance the accuracy and robustness of 3D reconstruction from multiple camera views.

Main Methods:

• Developed a common-view-based dual-view reconstruction method to minimize coordinate transformations.
• Constructed an image network using a baseline threshold method for high-quality image grouping.
• Introduced sum or reprojection residual of common points to validate orientation solutions.

Main Results:

• The proposed method effectively controls accumulative errors in the 3D reconstruction process.
• Experiments with synthetic and real images confirmed robust and highly accurate 3D shape acquisition.
• The common-view strategy significantly reduces error propagation compared to traditional methods.

Conclusions:

• The common-view-based dual-view reconstruction offers a robust solution for accurate large-scale 3D shape acquisition.
• This method addresses the critical challenge of accumulative errors in multi-view 3D reconstruction.
• The findings have implications for fields requiring precise 3D modeling from image data.