RANSAC算法的c++实现版本

Author: Ziv Yaniv (zivy@isis.georgetown.edu) This is an updated version of the original RANSAC code posted on my web site. Originally implemented circa 2000. Project implementation included a Makefile for unix systems and a VC++ 6.0 project for windows systems. Currently, 2010, I manage multi-platform code using the CMake (http://www.cmake.org) configuration tool. There were also changes to the file names and code. The code is now published under a BSD license (see license.txt). This archive contains an implementation of the Random Sample Consensus (RANSAC) framework for robust parameter estimation. The code comes in two flavors, a static function call approach (RANSAC.h) and a multi-threaded approach (RANSACMT.h). The former only requires that the user invoke a static function and pass it the relevant arguments while the latter requires object instantiation and optionally setting the number of threads to use. The archive includes the following files: 1. Cmakelists.txt - input file for the Cmake program which will generate platform specific configuration files (Makefile on linux/unix, solution file for visual studio on windows...). 2. Source files: Point2D.{cxx,h} - Primitive 2D point class used in the example. ParameterEstimator.h - A virtual class, the ancestor of all parameter estimators. LineParamEstimator.{cxx,h} - Class which implements the virtual functions defined by the ParameterEstimator class, for 2D lines. RANSAC.{h,txx} - Template implementation of the RANSAC algorithm using a static function approach. RANSACMT.{h,txx} - Template implementation of the RANSAC algorithm using a multi threaded approach. main.cxx - An example which shows how to use the RANSAC static function implementation. mainMT.cxx - An example which shows how to use the multi threaded RANSAC implementation. By default the number of threads is equivalent to the number of cores on the machine. To manually set this value uncomment lines 89-90 and set the appropriate value. multiThreading directory - A platform/OS independent library for multi-threading. Implementation appropriated with minor modifications from the Insight Registration and Segmentation toolkit (ITK - www.itk.org). Notes: 1. The math underlying the code which performs a least squares fit for the line is described in the summary on Principle Component Analysis (PCA) found on my web site (look under lesson handouts). Read the orthogonal least squares section and the section about eigenvectors/eigenvalues of symmetric 2X2 matrices. 2. For a detailed description of the RANSAC algorithm see: * Martin A. Fischler, Robert C. Bolles, ``Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography'', Communications of the ACM, Vol. 24(6), 1981. **Richard I. Hartely, Andrew Zisserman, "Multiple View Geometry in Computer Vision", Cambridge University Press, 2000. 3. This template implementation has been successfully used to estimate the parameters of: a. lines b. circles c. ellipses d. homographies This archive is intended to be self contained. As a consequence it only includes the 2D line parameter estimator as it is the only estimator that does not require the availability of linear algebra packages or non-linear optimization code.