Hough Transform and Clustering for a 3-D Building Reconstruction with Tomographic SAR Point Clouds

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Computer Vision

Background:

• Tomographic Synthetic Aperture Radar (TomoSAR) generates 3-D point clouds for urban areas.
• TomoSAR data often contains noise and false targets, degrading 3-D image quality and building reconstruction.
• Existing Hough transform methods for building outline detection produce fragmented lines and redundant parameters.

Purpose of the Study:

• To develop an improved method for detecting and refining building outlines from TomoSAR point clouds.
• To enhance the accuracy of 3-D building reconstruction by mitigating noise and false targets.
• To integrate fragmented line segments into continuous building outlines.

Main Methods:

• Utilized Hough transform to detect initial building outline segments in TomoSAR point clouds.
• Employed unsupervised clustering to group similar line parameters, representing continuous building segments.
• Integrated repaired line segments and estimated joint points to form complete building outlines.
• Applied supervised learning for refining building point clouds using the estimated outlines.
• Fed refined and unrefined data into a neural network for 3-D construction.

Main Results:

• Successfully integrated fragmented line segments into continuous building outlines.
• Significantly reduced noise and false targets in TomoSAR point clouds.
• Demonstrated improved accuracy and effectiveness in 3-D building reconstruction compared to existing methods.

Conclusions:

• The proposed method effectively refines building outlines from TomoSAR data.
• The enhanced outlines lead to higher quality 3-D building reconstructions.
• This approach offers a robust solution for urban area analysis using TomoSAR.