A Vision-Guided Point Cloud Refinement Method for Helical Surface Measurement: A Case Study on Screw Thread Pitch

Abstract

Screws are critical mechanical components, and their geometric accuracy directly determines the performance and reliability of mechanical systems. However, conventional contact-based inspection methods are time-consuming and limited to sparse sampling. Existing non-contact optical techniques often suffer from insufficient edge localization accuracy due to discrete point spacing. To address these challenges, this paper proposes a vision-guided point cloud refinement measurement method. The system integrates a laser scanner with a camera, where the laser scanning serves as the primary channel for acquiring three-dimensional point clouds, and the vision system provides high-resolution thread edge positions through sub-pixel localization and interpolation. A joint 2D-3D registration procedure is established to transfer the 2D sub-pixel edge localization precision to the corresponding 3D points, effectively correcting the spatial positions of thread crests. A coarse-to-fine processing pipeline is then applied, encompassing PCA-based principal axis alignment, Fourier transform-based coarse pitch estimation, and Gaussian pyramid-based multi-resolution refinement, to extract the pitch from the refined point cloud. Experimental validation on a real screw specimen confirms that the proposed method achieves a pitch measurement error of 0.0423 mm, demonstrating performance comparable to the pure vision-based method. It can provide a reliable solution for high-precision screw thread inspection.