Stereo vision system on machine tool for automated reconstruction of surface morphology with depth discontinuity

This paper describes a stereo algorithm based on a new variable window for surface morphology inspection and its reconstruction on a machine tool. It has been surveyed that previous stereo algorithms generate large matching errors due to large disparity variations near depth discontinuities. The proposed stereo algorithm devises the variable window in three-dimensional disparity space constructed with match candidate points, which are selected by a correlation function with robust estimator, including line masks. Also, it takes into account intensity variations using intensity gradient-based similarity and then reflects on projective distortion in the variable window to estimate accurate and robust disparities in surface morphology with depth discontinuity. Several experiments show that the proposed technique outperforms closely related stereo methods.