Efficient 2D and 3D Facade Segmentation Using Auto-Context.

Area of Science:

• Computer Vision
• Machine Learning
• 3D Reconstruction

Background:

• Building facade segmentation is crucial for urban modeling and analysis.
• Existing methods often rely on domain-specific information, limiting their generalizability.
• There is a need for efficient and robust segmentation techniques applicable to both 2D images and 3D point clouds.

Purpose of the Study:

• To introduce a novel, domain-independent segmentation technique for building facades.
• To evaluate the performance of the proposed method against existing approaches on benchmark datasets.
• To demonstrate the efficiency and extensibility of the developed system.

Main Methods:

• Utilizing a sequence of boosted decision trees trained with auto-context features.
• Employing stacked generalization for feature learning.
• Applying standard segmentation methods for a domain-independent approach.

Main Results:

• The proposed method achieves performance comparable to or better than previous published methods.
• Empirical results validated on all available 2D and 3D facade benchmark datasets.
• Demonstrated superior efficiency in test-time inference.

Conclusions:

• The developed segmentation technique is effective and efficient for building facades.
• The method's domain-independent nature and reliance on standard techniques enhance its applicability.
• The system is simple to implement and extend, offering practical advantages for researchers and developers.