An extended and optimized implementation of the state-of-theart
local curve fitting algorithm named Contracting Curve Density (CCD) algorithm, originally
developed by Hanek et al.
The CCD algorithm can be best described as follows. Given one or multiple images as
input data and a parametric curve model with a priori distribution of model parameters,
through curve-fitting process, we estimate the model parameters which determine the approximation
of the posterior distribution in order to make the curve models best matching
the image data. In order to improve the stability, accuracy and robustness over the original
implementation we introduce the following improvements. Firstly, we use the logistic
sigmoid function instead of a Gaussian error function which renders a curve-fitting problem
as a Gaussian logistic regression problem known in the field of pattern recognition.
Secondly, a quadratic or a cubic B-spline curve is used to model the parametric curve to
avoid the Runge phenomenon without increasing the degree of the B-spline. Thirdly, the
system supports both planar affine (6-DOF) and three-dimensional affine (8-DOF) shapespace.
The latter affine space can avoid curve mismatching caused by major viewpoint
changes. Lastly, in order to avoid manual intervention by the user, the developed system
also supports robust global initial curve initialization modules based on both keypoint
feature matching and back-projections from the 3D point clouds.
ccd is ...