Integrated Change Detection and Classification in Urban Areas Based on Airborne Laser Scanning Point Clouds

Area of Science:

• Geospatial Science
• Computer Science
• Remote Sensing

Background:

• Change detection in 3D point clouds is crucial for monitoring environmental and urban dynamics.
• Existing methods often involve separate classification and change detection steps, increasing complexity.
• Accurate analysis of 3D point cloud data requires robust feature extraction and processing techniques.

Purpose of the Study:

• To propose a novel, integrated approach for change detection (CD) in 3D point clouds.
• To combine point cloud classification and CD into a single machine learning-based step.
• To develop a method achieving high accuracy in identifying various types of changes.

Main Methods:

• Merged point cloud data from two epochs for feature computation.
• Utilized four types of features: point distribution, relative terrain elevation, laser scanning specific features, and cross-epoch features.
• Employed supervised classification based on acquired training samples for change identification.

Main Results:

• Achieved an overall accuracy exceeding 90% for both epochs.
• Successfully identified eight distinct classes of change, including lost/new trees and buildings, and ground changes.
• Demonstrated the effectiveness of the integrated approach for detailed change analysis.

Conclusions:

• The proposed integrated approach offers a highly accurate and efficient solution for 3D point cloud change detection.
• The method's ability to combine classification and CD simplifies the analysis pipeline.
• This technique provides valuable insights for environmental and urban change monitoring.