Combined image block and DBSCAN processing for large field of view structured light point cloud denoising

Area of Science:

• Computer Vision
• 3D Reconstruction
• Image Processing

Background:

• Stripe projection structured light systems are crucial for 3D reconstruction in industrial inspection and virtual reality.
• Increasing demands for large field-of-view reconstruction amplify environmental noise, degrading point cloud quality.
• Existing noise reduction methods may not adequately preserve detail in complex 3D data.

Purpose of the Study:

• To develop a robust combined preprocessing method for 3D point cloud data.
• To effectively reduce noise and outliers in large field-of-view 3D reconstructions.
• To improve the accuracy and completeness of reconstructed 3D morphology.

Main Methods:

• A combined preprocessing approach using phase field and binocular matching data.
• Image block statistical analysis for outlier detection in phase field data.
• DBSCAN clustering on binocular matching points for outlier rejection.
• Integration with camera calibration for final 3D shape reconstruction.

Main Results:

• The proposed method demonstrates superior noise point removal compared to Gaussian, bilateral, and statistical filtering.
• It effectively retains target point cloud data, preserving intricate details.
• Significant improvement in the quality of 3D point cloud reconstruction for large fields of view was achieved.

Conclusions:

• The combined preprocessing method offers a more effective solution for noise reduction in 3D point cloud data.
• This technique enhances the fidelity of 3D morphology reconstruction, particularly for large-scale applications.
• The approach provides a valuable tool for improving the reliability of structured light systems.