Robust adaptive sub-pixel advancement extraction algorithm for line laser centers with curvature variation

Area of Science:

• Computer Vision
• Image Processing
• Computational Geometry

Background:

• Existing line-structured light center extraction algorithms struggle with missing center points, positioning deviations, and poor continuity, especially under complex curvature, stripe ends, and noise.
• These limitations hinder accurate 3D reconstruction and analysis in various machine vision applications.

Purpose of the Study:

• To propose a robust adaptive sub-pixel extraction algorithm for line laser centers that overcomes the limitations of existing methods.
• To enhance the accuracy, continuity, and completeness of laser stripe center extraction in challenging imaging conditions.

Main Methods:

• An adaptive sub-pixel extraction algorithm employing edge padding, structure tensor estimation for orientation field generation, and an internal advancement strategy.
• Utilizes bilinear interpolation along the normal direction and the gray-scale centroid method for high-precision sub-pixel localization.

Main Results:

• The algorithm demonstrates superior noise resistance, extracting all 401 center points at a noise variance of 0.14 with a Mean Absolute Error (MAE) below 0.23.
• Achieves continuous, complete, smooth, and redundant-free extraction of laser stripe centers across varying curvatures in real-world images.
• Outperforms classical algorithms like Steger and Directional Template Method (DTM) in accuracy and robustness.

Conclusions:

• The proposed algorithm offers a robust and efficient solution for line laser center extraction, significantly improving upon existing methods.
• It provides high-precision sub-pixel localization and maintains performance under complex conditions, making it suitable for demanding machine vision tasks.