Accurate multiple view 3D reconstruction using patch-based stereo for large-scale scenes.

Area of Science:

• Computer Vision
• Photogrammetry
• 3D Reconstruction

Background:

• Multiple View Stereo (MVS) methods are crucial for 3D scene reconstruction.
• Existing MVS methods often face challenges with large-scale scenes regarding accuracy and computational efficiency.

Purpose of the Study:

• To propose a novel depth-map merging based MVS method for large-scale scenes.
• To achieve both high accuracy and computational efficiency in 3D reconstruction.

Main Methods:

• An efficient patch-based stereo matching process generates initial depth maps for each image.
• A subsequent depth-map refinement step enforces consistency across neighboring views.
• The method is designed for parallel processing at the image level.

Main Results:

• The proposed method reconstructs accurate and dense point clouds.
• It demonstrates high computational efficiency compared to state-of-the-art techniques.
• Successfully evaluated on benchmark datasets and large-scale real-world data.

Conclusions:

• The developed depth-map merging MVS method offers a viable solution for large-scale 3D scene reconstruction.
• Its parallelizable nature and efficiency make it suitable for high-resolution imagery.
• The method balances accuracy and speed effectively.