No abstract

No abstract

The paper deals with the metrological characterization of provided. In absence of it no reliability universally assessed a stereo-vision based measurement system for the inspection of can be attributed to measured data, neither a serious automotive rubber profiles in an industrial plant. The comparison of the stated result can be made with a reference characterization of the class of such measurement systems value or other measurement results of the same physical introduces new challenges, both for the unavailability of reference quantity. Then efforts are justified to address the issue of measurement instruments andfor the complexity ofthe measurement giving methods that provides a quantitative indication of the system itselfthat does not allow a straightforward application ofthe standard procedures for the uncertainty evaluation. After a uncertainty of the results. This task could be accomplished description of the measurement system, the characterization is through suitable parameters. The accuracy is frequently presented in terms of evaluation of the dependencies of systematic adopted for characterizing an optical measuring system effects and uncertainties on known and expected influence although its definition is not universally accepted. As quantities. With this aim, several experimental results are reported example it can be defined either as the maximum among [9] and commented.or as the standard deviation of [10] the displacements Keywords -Contact-less measurement, machine vision, automotive registered between the nominal 3D world-coordinates (x or y profile, uncertainty evaluation, image processing or z) of test points of some non-coplanar objects and their reconstruction from measured data. Finally the evaluation of

This paper deals with the problems in setting up stereo-vision systems for contactless measurement of dimensional parameters in industrial environments. Two implicit calibration algorithms for the reconstruction of three-dimensional (3-D) real-world coordinates of objects from pairs of two-dimensional image coordinates have been implemented and compared. The former is based on a direct linear transformation, while the latter on an Artificial Neural Network (ANN). The results of the comparison made on artificial and real objects are finally reported in terms of statistical analysis of the reconstruction error

No abstract