A featureless approach to 3D polyhedral building modeling from aerial images

Area of Science:

• Computer Vision
• Photogrammetry
• 3D Reconstruction

Background:

• Reconstructing 3D building models from aerial imagery is crucial for urban planning and geospatial analysis.
• Existing feature-based methods often struggle with textureless regions and require complex matching steps.

Purpose of the Study:

• To develop a featureless, model-based approach for accurate 3D polyhedral building reconstruction from calibrated aerial images.
• To leverage direct optimization on raw image brightness, bypassing traditional feature extraction and matching.

Main Methods:

• Exploits geometric and photometric properties of perspective projection of planar structures.
• Employs direct optimization of an objective function combining image dissimilarity and gradient scores.
• Utilizes the Differential Evolution algorithm for the optimization process.

Main Results:

• Demonstrates accurate 3D polyhedral model reconstruction without feature extraction or matching.
• Achieves higher accuracy compared to conventional feature-based 3D reconstruction approaches.
• Validates feasibility and robustness through evaluations on real and synthetic aerial images.

Conclusions:

• The proposed featureless, model-based approach offers a robust and accurate method for 3D building reconstruction from aerial images.
• Direct optimization on raw image brightness provides an effective alternative to feature-based techniques.
• Enables fast 3D model rectification and updating, with significant implications for geospatial applications.