Page 296 - DMTH402_COMPLEX_ANALYSIS_AND_DIFFERENTIAL_GEOMETRY
P. 296

Unit 23: Developable Surface Fitting to Point Clouds




          23.5.4 Fitting Developable Surfaces to nearly Developable Shapes                      Notes

          The proposed method can be applied also to fit a developable surface to data which comes from
          a nearly developable shape. Of course, we have to specify what nearly developable means in
          this context. Since the fitting is performed by fitting a one-parameter family of tangent planes,
          we will formulate the requirements on the data pi in terms of the Blaschke image of the estimated
          tangent planes T . i
          If the data points p  are measurements of a developable surface D and if the width in direction of
                         i
          the generators does not vary  too much,  the Blaschke  image  b(D) =  R  will be a  tubular-like
          (curve-like) region on B with nearly constant thickness. Its boundary looks like a pipe surface.
          Putting small distortions to D, the normals of D will have a larger variation near these distortions.
          The Blaschke image b(D) possesses a larger width locally and will look like a canal surface. As
          long as it is still possible to compute a fitting curve to b(D), we can run the algorithm and obtain
          a developable surface approximating D. Figures 23.5 and 23.7 illustrate the projection of b(D)
          onto the unit sphere S .
                            2
          The  analysis  of  the  Blaschke  image  b(D)  gives  a  possibility  to  check  whether  D  can  be
          approximated by a developable surface or not. By using the cell structure of the cleaned Blaschke
          image b(D), we pick a cell C and an appropriately chosen neighborhood U of C. Forming the
          intersection R = U  b(D), we compute the ellipsoid of inertia (or a principal component analysis)
          of R. The existence of one significantly larger eigenvalue indicates that R can be approximated
          by a curve in a stable way. Thus, the point set corresponding to R can be fitted by a developable
          surface.
          Since approximations of nearly developable shapes by developable surfaces are quite useful for
          practical purposes, this topic will be investigated in more detail in the future.

          23.5.5 Singular Points of a Developable Surface

          So far, we did not pay any attention to singular points of D*. The control and avoidance of the
          singular points within the domain of interest is a complicated topic because the integration of
          this into the curve fitting is quite difficult.
          If the  developable surface  D* is  given by  a point  representation,  formula (4)  represents  the
          singular curve s(t). If D* is given by its tangent planes E(t), the singular curve s(t) is the envelope
          of the generators L(t) and so it is computed by

                                        
                                   
                         s(t) = E(t)   E(t)  E(t).                               (14)
          Thus, the singular curve s(t) depends in a highly nonlinear way on the coordinate functions of
          E(t).






















                                           LOVELY PROFESSIONAL UNIVERSITY                                  289
   291   292   293   294   295   296   297   298   299   300   301