An adaptive 3D point-cloud measurement framework for contour reconstruction of deformable rolled objects

Abstract

Accurate contour measurement of deformable rolled objects supports automated loading, dimensional inspection, and robotic handling but is degraded by deformation, clutter, ground interference, and depth noise. This paper organizes established operators into an uncertainty-oriented preprocessing chain and couples geometric and spatial evidence with an FEA-derived regional deformation prior. Under the selected settings, non-ground retention, outlier removal, and background removal reached 94.6%, 96.7%, and 93.2%, respectively. Reconstruction achieved 95.7% precision, 93.6% recall, and a 0.37 cm contour fitting error. Across the tested 3, 5, and 10 cm fabric-roll specifications, precision was 95.0%–96.3%, recall was 93.0%–94.8%, and fitting error was 0.34–0.42 cm. Controlled tests showed progressive degradation with deformation, background, noise/outliers, and density. The results support robustness within the tested fabric-roll class, without claims beyond it.