Automatic Indoor Reconstruction from Point Clouds in Multi-room Environments with Curved Walls

Area of Science:

• Computer Vision
• 3D Reconstruction
• Computational Geometry

Background:

• Advancements in laser scanning drive interest in indoor modeling.
• Semantically rich 3D indoor models are crucial for various applications.
• Existing 3D indoor reconstruction methods struggle with multi-room environments featuring curved walls.

Purpose of the Study:

• To develop a novel method for accurate 3D indoor modeling of environments with straight and curved walls.
• To address limitations in current point cloud-based indoor reconstruction techniques.

Main Methods:

• A novel straight and curved line tracking method with a straight line test.
• Constrained least squares for straight line regularization.
• Cell complex construction integrating straight and curved lines.
• Markov Random Field formulation for indoor reconstruction as a labeling problem.
• Minimum graph cut approach to minimize energy function for optimal labeling.

Main Results:

• The proposed method successfully reconstructs multi-room indoor environments with curved walls.
• Demonstrated robustness in handling complex geometric features.
• Achieved accurate and semantically rich 3D indoor models.

Conclusions:

• The developed method is well-suited for 3D indoor modeling, particularly in complex spaces with curved walls.
• Offers a significant improvement over existing techniques for challenging indoor environments.
• Enables creation of detailed and accurate 3D indoor models from point cloud data.