Single-shot multi-line structured light stripe recognition based on deep learning.

Area of Science:

• Optics and Photonics
• Computer Vision
• Metrology

Background:

• Multi-line structured light measurement is crucial for high-speed, high-precision 3D surface profiling.
• Challenges include stripe cracks and misalignments, hindering sequential numbering.
• Existing methods often require complex auxiliary patterns, reducing efficiency.

Purpose of the Study:

• To develop an efficient and accurate multi-line structured light stripe numbering method.
• To overcome limitations of traditional methods relying on auxiliary encoding patterns.
• To enable robust stripe numbering without additional hardware.

Main Methods:

• Utilized deep learning for semantic segmentation of multi-line structured light stripes.
• Applied center line extraction to segmented stripes for ordering and numbering.
• Developed a method requiring only a single multi-line stripes pattern.

Main Results:

• Successfully performed semantic segmentation of stripes without auxiliary patterns.
• Accurately determined stripe ordering and numbering through center line analysis.
• Validated the method's feasibility in complex scenarios, including high reflectivity and scattering.

Conclusions:

• The proposed deep learning method offers an efficient and flexible solution for multi-line structured light stripe numbering.
• Eliminates the need for projected auxiliary encoding patterns, simplifying the setup.
• Demonstrates effectiveness in challenging measurement environments, advancing 3D surface profiling techniques.