A Laser Scanning-Based Centerline Extraction Algorithm Based on Grayscale and Curvature Constraints for Three-Dimensional Shape Measurement

In the field of industrial nondestructive testing (NDT), the accurate acquisition of three-dimensional (3D) shape measurement is crucial for defect identification and structural assessment. As an important branch of 3D shape measurement techniques, laser scanning technology plays a key role in machine vision due to its high precision and noncontact nature. In laser scanning technology, the accuracy of 3D coordinate calculation depends on rapid high-precision extraction of laser stripe centerlines. To address this, this paper proposes a grayscale-curvature constrained algorithm to enhance the speed and accuracy of laser stripe centerline extraction. The process employs median filtering and improved anisotropic gaussian filtering (IAGF) for noise suppression. An energy function combining grayscale and curvature information is then defined. For each image row, the pixel minimizing this energy function is selected as the center candidate. Finally, subpixel refinement is performed along the normal direction based on the gradient-tangent orthogonality principle. Experimental results demonstrate that IAGF reduces centerline extraction error by 28.7% compared to standard gaussian filtering. Furthermore, the proposed algorithm significantly outperforms the grayscale centroid, normal-based grayscale centroid, and conventional Steger methods, showing accuracy improvements of 58.3, 46.2, and 35.4% respectively. The reconstructed dimensions of a gauge block exhibit minimal relative errors, specifically only 0.04% for length, 0.07% for width, and 0.17% for thickness. The proposed algorithm enhances 3D shape measurement precision, which can be effectively applied in the field of NDT industrial applications.