A discussion about stereo vision techniques for industrial image-based measurement systems

Abstract

AAt -paper deals with theproblems in setting up stereo-vision systems for contort-less measurement of dimensional parameters in industrial environments. Two implicit calibration algorithmsfor fhe reconstruction of 30 real world coordinafes of objects from pairs of 2D image coordinates have been implemented and compared The former is based on a direct linear transformation, while the Inner on an Arfificial Neural Nefwork The results of a not exhaustive comparison, made on arti/ciaI objecfs, are finally reported in terms of statisfical analysis of the reconstruction error.