HDRSL Net for Accurate High Dynamic Range Imaging-Based Structured Light 3D Reconstruction.

Area of Science:

• Optics and Photonics
• Computer Vision
• Metrology

Background:

• Accurate 3D reconstruction of objects with varying surface reflectivity, particularly in fringe projection profilometry, necessitates high dynamic range (HDR) imaging.
• Single-exposure cameras struggle to capture HDR fringe patterns efficiently due to limited dynamic range, hindering precise 3D measurements.
• Existing methods often require multiple exposures or complex setups, increasing acquisition time and system complexity.

Purpose of the Study:

• To develop an efficient and accurate deep learning-based pipeline for HDR structured light 3D reconstruction.
• To address the limitations of single-exposure cameras in capturing HDR fringe patterns for 3D measurement.
• To enable robust and precise 3D reconstruction of objects with challenging HDR reflective surfaces.

Main Methods:

• An end-to-end deep learning pipeline comprising an HDR Fringe Generation Module and a Phase Calculation Module was developed.
• The HDR Fringe Generation Module reconstructs HDR fringe images from short- and long-exposure low dynamic range (LDR) inputs using attention guidance and feature distillation.
• The Phase Calculation Module processes phase information from the generated HDR fringes for 3D reconstruction.

Main Results:

• The proposed method achieved a phase error of 0.105 on a metallic HDR dataset, comparable to multi-exposure Phase Shifting Profilometry (PSP) but with significantly reduced projection time (8.3%).
• Quantitative measurements demonstrated sub-50 $\mu $ m accuracy on various objects, including ceramic spheres, flat plates, and metal steps.
• Ablation studies confirmed the effectiveness of feature distillation and attention mechanisms in generating high-quality HDR fringe patterns.
• A new HDR imaging metal dataset with 1,700 samples was created as a benchmark for HDR structured light measurement.

Conclusions:

• The deep learning pipeline provides an efficient and generalizable solution for HDR structured light 3D reconstruction.
• The method demonstrates robustness across diverse object geometries, exposure levels, and challenging global illumination conditions.
• The developed approach significantly improves the efficiency and accuracy of 3D reconstruction for objects with HDR reflective surfaces.