Non-invasive multilevel geometric regularization of mesh-based 3D shape measurement

International audienceFinite element stereo digital image correlation (FE-SDIC) requires a crucial calibration phase in which the initial CAD needs to be updated to fit the actual shape of the specimen. On the one hand, the use of a FE mesh facilitates the coupling of measurements with simulation tools. On the other hand, it provides a unique, fine description of both the geometry and the displacement, which often makes the shape measurement problem highly ill-posed. As a remedy, we propose a hybrid isogeometric-FE strategy that can measure a shape in terms of spline functions while considering as an input and output the analysis-suitable FE mesh. Making use of the appealing spline refinement procedures and of Bézier-based operators, multilevel smooth spline dis-cretizations are built concurrently with the initial FE subspace and related to the multi-scale images used for the initialization of the shape measurement. It results in a geometrically sound regularization which provides a spline parametrization of the optimal shape along with its FE twin. A non-invasive implementation from an existing FE-SDIC code is also detailed. The performance of the proposed method is assessed on real images and comparisons are made with other published techniques to prove its efficiency