Robust 3D reconstruction in noisy environments

Abstract

Automated inspection in industrial manufacturing can minimize the total production cost of a part. Current inspection solutions often involve measuring a part manually, which interrupts the machining process. We present two non-contact real-time systems which integrate visual inspection in-line with CNC (computer numerical control) machines and ensure dimensional model generation of parts with high accuracy. We first present a camera-projector scanning system that uses photometric stereo and structured light scanning to reconstruct the shape of objects in the presence of specular chip-like noise and high-speed object revolution. We obtain reconstruction accuracies down to 0.5 mm for objects with complex reflectance on a representative CNC lathe. For rotationally symmetric objects, we also propose a novel shape from silhouette system which uses principles from light transport theory to efficiently image transmissive paths through a scattering medium. The system enables in-line and highly accurate geometric reconstructions down to 60 μm on CNC lathe machines in the presence of scattering fluid and specular metallic shavings. Both systems are compact and cost-effective alternatives to the current use of CMMs (coordinate measuring machines) for manual inspection of machined parts.