A Robust Gradient Based Method for Building Extraction from LiDAR and Photogrammetric Imagery.

Area of Science:

• Geospatial analysis
• Remote sensing technology
• Computer vision for urban mapping

Background:

• Current automatic building extraction methods struggle with small buildings and transparent roofs.
• Large area thresholds and ground point usage limit detection accuracy.
• Complex environments and numerous parameters reduce the robustness of existing methods.

Purpose of the Study:

• To propose a novel Gradient-based Building Extraction (GBE) method.
• To address limitations in extracting small, transparent-roofed buildings, and improve robustness in complex terrains.
• To develop an automated method with minimal manual parameter tuning.

Main Methods:

• Transforming LiDAR height data into an intensity image without height interpolation.
• Analyzing image gradients to differentiate building roofs from vegetation.
• Employing local color matching, variance, point density, and shadow elimination for post-processing.
• Automating parameter settings from the data, minimizing manual intervention.

Main Results:

• Successfully extracted buildings of various sizes, including small ones.
• Effectively detected buildings with both transparent and opaque roofs.
• Demonstrated superior performance in eliminating trees and handling complex environments.
• Outperformed existing state-of-the-art methods in object and pixel-based evaluation metrics.

Conclusions:

• The Gradient-based Building Extraction (GBE) method offers a robust solution for automatic building extraction.
• GBE overcomes limitations of existing methods, improving accuracy for diverse building types and environments.
• The automated approach enhances the reliability and applicability of building extraction from LiDAR data.