DEVELOPMENT OF PREPROCESSING METHOD FOR RECONSTRUCTION OF GEOMETRIC 3D OBJECT BASED ON COLOR-FRINGE PATTERN PROFILOMETRY

• Main Author: Ginanjar, Teguh
• Format: Theses
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/48074
• Institution: Institut Teknologi Bandung

Fringe projection is an advantageous technique for the acquisition of high-resolution, real-time three-dimensional surface data measurement due to it is non-invasive and non-contact properties. By employing an RGB colour fringe, a deformation of the surface object can be measured with a minimum problem with motion artefact. However, the colour fringe projection introduced uncertainties in 3D surface reconstruction due to colour unbalance and colour crosstalk. Thus, a preprocessing scheme of phase image proposed to correct undesired 3D surface. The source of errors of colour fringe projection assumes due to the nonlinearity of projection intensity, background noise, surface reflectivity, colour imbalance and colour crosstalk. Fringe images were simulated with induced noise as follows, (i) nonlinearity model with power factor of 2.2; (ii) reflectivity factor between 50%-60%; (iii) color imbalance with 90-94% from original value; (iv) color crosstalk with factor of 5% for each crosstalk between color channel and (v) random noise with factor of 5%. The proposed preprocessing scheme is consisting of some steps, i.e. (i). Nonlinearity intensity correction using a simple parametric curve fitting; (ii) Filtering technique and (iii). Normalization of fringe amplitude. Furthermore, to surface profile smoothing with keep fidelity of curve, mean curvature diffusion (MCD) filter performed on the output of the absolute phase image. Reconstruction 3D profile of four simulated objects (i.e. symmetry dome, one peak of Gaussian surface, three peaks of Gaussian surface and two peaks of prism surface) has mean of errors in about 40% relative to the desired 3D profile of those object. The proposed scheme can recover the 3D profile of four objects with errors of less than 6% than the desired 3D profile. By application of MCD filter on post-processing, the better recover the 3D profile can be achieved with mean errors reduce to 4%.