next up previous contents
Next: 7.5 Cross-Sectional Design Up: 7 Modeling Environment and Previous: 7.3 Solid Rounding

7.4 Optimal Surface Fitting

 

D-NURBS are applicable to the optimal fitting of regular or scattered data [28]. The most general and often most useful case occurs with scattered data, when there are fewer or more data points than unknowns--i.e., when the solution is underdetermined or overdetermined by the data. In this case, D-NURBS can yield ``optimal'' solutions by minimizing the thin-plate under tension deformation energy [35, 33]. The surfaces are optimal in the sense that they provide the smoothest curve or surface (as measured by the deformation energy) which interpolates or approximates the data.

The data point interpolation problem amounts to a linear constraint problem when the weights tex2html_wrap_inline1967 are fixed, and it is amenable to the constraint techniques presented in Section 5.2. The optimal approximation problem can be approached in physical terms, by coupling the D-NURBS to the data through Hookean spring forces (19). We interpret tex2html_wrap_inline2059 in (19) as the data point (generally in tex2html_wrap_inline2385 ) and tex2html_wrap_inline2057 as the D-NURBS parametric coordinates associated with the data point (which may be the nearest material point to the data point). The spring constant c determines the closeness of fit to the data point.gif

We present three examples of surface fitting using D-NURBS coupled to data points through spring forces. Fig. 3(a) shows 19 data points sampled from a hemisphere and their interpolation with a quadratic D-NURBS surface with 49 control points. Fig. 3(b) shows 19 data points and the reconstruction of the implied convex/concave surface by a quadratic D-NURBS with 49 control points. The spring forces associated with the data points are applied to the nearest points on the surface. In Fig. 3(c) we reconstruct a wave shape from 25 sample points using springs with fixed attachments to a quadratic D-NURBS surface with 25 control points.

 

  figure620


Figure 3: Optimal surface fitting: D-NURBS surfaces fit to sampled data from (a) a hemisphere, (b) a convex/concave surface, (c) a sinusoidal surface. (a-c1) D-NURBS patch outline with control points (white) and data points (red) shown. (a-c2) D-NURBS surface at equilibrium fitted to scattered data points. Red line segments in (c2) represent springs with fixed attachment points on surface.


next up previous contents
Next: 7.5 Cross-Sectional Design Up: 7 Modeling Environment and Previous: 7.3 Solid Rounding

Demetri Terzopoulos | Source Reference