How to select good neighboring images in depth-map merging based 3D modeling

Area of Science:

• Computer Vision
• 3D Reconstruction
• Computational Photography

Background:

• Depth-map merging is vital for large-scale 3D modeling from images.
• Selecting appropriate neighboring images significantly impacts reconstruction quality.
• This crucial step has been under-addressed in existing literature.

Purpose of the Study:

• To address the challenge of selecting suitable neighboring images for large-scale 3D scene reconstruction.
• To improve the quality of depth maps and final 3D models.
• To handle reconstructions with numerous unordered images, varying scales, and view-angle changes.

Main Methods:

• Formulating neighboring image selection as a combinatorial optimization problem.
• Employing a quantum-inspired evolutionary algorithm to find optimal solutions.
• Evaluating the approach on ground truth datasets for large-scale scene reconstruction.

Main Results:

• Demonstrated significant improvements in depth map quality.
• Achieved enhanced accuracy in the final 3D reconstruction results.
• Showcased high computational efficiency of the proposed method.

Conclusions:

• The proposed method effectively tackles the neighboring image selection problem in 3D reconstruction.
• Quantum-inspired evolutionary algorithms offer a powerful tool for optimizing this process.
• This approach leads to higher quality large-scale 3D models with improved efficiency.