Fast B-Spline 2D Curve Fitting for unorganized Noisy Datasets

Nadim El-Hayek (1) , Olivier Gibaru (1) , M. Damak (1, 2) , H. Nouira (3) , Nabil Anwer (4) , Eric Nyiri (5)

1 LSIS - Laboratoire des Sciences de l'Information et des Systèmes
2 Geomnia
3 FEMTO-ST - Franche-Comté Électronique Mécanique, Thermique et Optique - Sciences et Technologies (UMR 6174)
4 LURPA - Laboratoire Universitaire de Recherche en Production Automatisée
5 LSIS- INSM - Laboratoire des Sciences de l'Information et des Systèmes : Ingénierie Numérique des Systèmes Mécaniques

In the context of coordinate metrology and reverse engineering, freeform curve reconstruction from unorganized data points still offers ways for improvement. Geometric convection is the process of fitting a closed shape, generally represented in the form of a periodic B-Spline model, to data points [WPL06]. This process should be robust to freeform shapes and convergence should be assured even in the presence of noise. The convection's starting point is a periodic B-Spline polygon defined by a finite number of control points that are distributed around the data points. The minimization of the sum of the squared distances separating the B-Spline curve and the points is done and translates into an adaptation of the shape of the curve, meaning that the control points are either inserted, removed or delocalized automatically depending on the accuracy of the fit. Computing distances is a computationally expensive step in which finding the projection of each of the data points requires the determination of location parameters along the curve. Zheng et al [ZBLW12] propose a minimization process in which location parameters and control points are calculated simultaneously. We propose a method in which we do not need to estimate location parameters, but rather compute topological distances that can be assimilated to the Hausdorff distances using a two-step association procedure. Instead of using the continuous representation of the B-Spline curve and having to solve for footpoints, we set the problem in discrete form by applying subdivision of the control polygon. This generates a discretization of the curve and establishes the link between the discrete point-to-curve distances and the position of the control points. The first step of the association process associates BSpline discrete points to data points and a segmentation of the cloud of points is done. The second step uses this segmentation to associate to each data point the nearest discrete BSpline segment. Results are presented for the fitting of turbine blades profiles and a thorough comparison between our approach and the existing methods is given [ZBLW12, WPL06, SKH98].

Mots clés: B-Spline; active contour deformation

Domaines: Génie mécanique [physics.class-ph]; Génie mécanique [physics.class-ph]

Nadim El-Hayek, Olivier Gibaru, M. Damak, H. Nouira, Nabil Anwer, et al.. Fast B-Spline 2D Curve Fitting for unorganized Noisy Datasets. 8th international conference on Curves and Surfaces, Jun 2014, Paris, France. ICCS 2014. ⟨hal-01069562⟩