Cross-grid laser reconstruction method based on stripe splitting and coarse-to-fine matching

Area of Science:

• Computer Vision
• Metrology
• Optical Engineering

Background:

• Conventional 3D reconstruction struggles with laser stripe center point (LSCP) matching accuracy due to laser plane deformations and increased laser stripes.
• Unidirectional laser stripe distribution in parallel multi-line systems complicates registration and prolongs scanning times.

Purpose of the Study:

• To introduce a novel trinocular cross-grid laser (CLS) reconstruction method to overcome limitations of conventional systems.
• To enhance the accuracy, efficiency, and robustness of multi-line laser 3D reconstruction.

Main Methods:

• System calibration to estimate trifocal tensor (TFT) and cross-grid laser (CLS) splitting reference angle.
• CLS splitting into unidirectional laser stripes (LaS) using Hessian-matrix normal vector and minimum directed boundary distance.
• Coarse LSCP matching via epipolar geometry, disparity, and spatial consistency, followed by fine matching using TFT-based point transfer.
• 3D reconstruction using left and right cameras and global registration of point clouds with retroreflective markers.

Main Results:

• Achieved a matching rate (MRa) of 97.4561% and matching accuracy rate (MAR) of 99.1072% in a 15-line CLS system.
• Reconstruction errors for 2D planes and 3D spheres were below 0.037 mm and 0.095 mm, respectively.
• The system operates at approximately 12 fps with GPU acceleration, demonstrating high efficiency.

Conclusions:

• The proposed CLS method significantly reduces reliance on the laser-plane equation for multi-line laser reconstruction.
• This study presents a novel and effective approach for line-laser 3D reconstruction, improving upon existing techniques.