Building Change Detection from Bi-Temporal Dense-Matching Point Clouds and Aerial Images

Area of Science:

• Geospatial analysis
• Photogrammetry
• Computer vision

Background:

• Accurate building change detection is crucial for urban planning and monitoring.
• Existing methods struggle with robustly identifying above-ground changes and classifying them.

Purpose of the Study:

• To develop a novel method for robust building change detection from bi-temporal point clouds and aerial images.
• To automatically classify detected changes into buildings and non-buildings.
• To categorize changes into specific types: newly built, taller, demolished, or lower.

Main Methods:

• Change detection framed as binary classification (foreground/background) using graph cuts on point clouds.
• Region-growing algorithm to form candidate changed building objects.
• Novel structural feature extraction from aerial images for building classification.
• Classification combined with digital surface models for detailed change categorization.

Main Results:

• The method effectively acquires changed objects above ground.
• Accurate classification of changed objects into buildings and non-buildings.
• Successful categorization of building changes into four distinct types.
• Validation on a large dataset demonstrates high effectiveness.

Conclusions:

• The proposed method offers a robust and automated solution for building change detection.
• It significantly improves the classification accuracy of changed objects.
• The approach is effective for monitoring urban development and changes over time.