A Rapid 3D Vision Inspection System for Sheet Metal Parts Based on Feature Extraction and Partial Point Clouds

Abstract

3D vision inspection of sheet metal parts is usually a time-consuming task and hence not suitable for real-time measurement of 3D geometric features.To address this problem, this work presents an active binocular 3D vision inspection system for rapid measurement of features of sheet metal parts using 2D feature extraction and partial 3D point cloud. The inspectionsystem firstly extracts the feature to be measured from one 2D image, then identifies the homologous points describing the feature on the other 2D image with the assistance of structured light, andat lastselectively computes the 3D coordinates related to thefeature measurement to largely reduce computational complexity. The system is applied to a sheet metal parts punching production line to monitor a couple of critical dimensionsin real time. The experiment results indicate that it canacquireaperture center coordinate, radius, and aperture distances within 0.1 seconds and with an average error of less than 0.2mm. The proposed system shows the potentialfor being deployed on mass-production lines to realize rapid 3D measurementfor metal sheet parts.