A scalable structure-from-motion framework for efficient 2D-to-3D reconstruction of historical artifacts

Abstract

Virtual museums are becoming increasingly popular, offering accessibility and engagement with historical artifacts for wider audiences. However, creating high-quality 3D models of artifacts can be time-consuming and expensive. Addressing these issues, we propose a comprehensive SFM based 2D-to-3D reconstruction method that enhances accessibility and scalability. Our approach integrates efficient camera calibration achieving up to 95% accuracy, robust feature detection and matching using algorithms such as SIFT, ORB, and AKAZE, and employs pose estimation, triangulation, and bundle adjustment to ensure high accuracy and detail. The method is lightweight, minimizing computational load, and is implemented on a user-friendly web-based platform. The solution demonstrated promising results, with reprojection errors as low as 15%, and effective 3D reconstructions of artifacts. Applications include virtual museums and the preservation and virtualization of artifacts, providing an interactive and immersive experience for users.