Inspection of continuously moving metallic tubes by artificial vision: modelisation and results

We present the problems faced in the design of an artificial vision inspection system of metallic tubes and the solutions adopted. We propose a theoretical model of image acquisition and discuss the signal processing problems in the case of artificial vision inspection of continuously moving products: the influence of acquisition period, resolution, movements and defect size on detection scattering; the signal-to-noise ratios are estimated. Finally, we present some results obtained on production line. Our solution is based on the use of a set of linear CCD cameras, ensuring the achievement of a 2D image. The tint defect can be detected by using two different view angles. The introduction of an angular correction reduces the scattering of the system responses. The calculation of the parameters in a low level language enables us to perform a real time inspection