Progressive Model-Driven Approach for 3D Modeling of Indoor Spaces

Area of Science:

• Computer Vision
• 3D Reconstruction
• Geomatics Engineering

Background:

• 3D indoor modeling is crucial for various applications.
• Three-dimensional point clouds from laser scanners are widely used for this purpose.
• Existing methods face challenges with noisy, cluttered, and obstructed data.

Purpose of the Study:

• To develop a robust method for 3D indoor space modeling.
• To address data imperfections like obstruction, clutter, and noise.
• To reconstruct detailed and accurate 3D models of building interiors.

Main Methods:

• A model-driven approach using watertight predefined models for indoor space reconstruction.
• A two-step process enabling the modeling of non-rectangular spaces.
• An improvement step to enhance model detail by incorporating intrusions and protrusions.
• 3D model generation through 2D to 3D extrusion.

Main Results:

• The algorithm successfully reconstructs indoor spaces from laser scan point clouds.
• Achieved completeness ranging from 77% to 95%.
• Achieved correctness ranging from 85% to 97%.
• Demonstrated geometric accuracy between 1.7 cm and 2.4 cm.

Conclusions:

• The proposed model-driven algorithm effectively reconstructs 3D indoor spaces.
• The method demonstrates high performance in completeness, correctness, and geometric accuracy.
• Validated with four benchmark real-world datasets, proving its efficacy.