Revisiting Projective Structure from Motion: A Robust and Efficient Incremental Solution.

Area of Science:

• Computer Vision
• 3D Reconstruction
• Computational Geometry

Background:

• Projective Structure from Motion (PSfM) is crucial for 3D scene reconstruction.
• Existing methods struggle with missing data, outliers, and large-scale scenarios.
• Efficiently estimating projective depths is a key challenge.

Purpose of the Study:

• To develop a robust and efficient PSfM solution for large-scale 3D reconstruction.
• To address challenges posed by missing data and outliers.
• To improve computational speed compared to traditional methods.

Main Methods:

• Embedding projective depths into projective parameters to reduce unknowns.
• Utilizing linear Least Squares optimization on standard systems instead of homogeneous ones.
• Employing an incremental approach with robust outlier-free procedures and refinement techniques.

Main Results:

• Demonstrated efficient handling of missing data and outliers in PSfM.
• Achieved significant improvements in computational speed.
• Validated performance on simulated and real-world datasets, including those with up to 98% missing data.

Conclusions:

• The proposed method offers a robust solution for large-scale 3D reconstruction using PSfM.
• It effectively manages significant amounts of missing data and outliers.
• The approach provides a computationally efficient alternative for complex reconstruction tasks.