SLAM.zip_SLAM初始源程序及学习教程_slam_slam算法学习_视觉SLAM

************************************************************** sparseLM version 1.3 By Manolis Lourakis Institute of Computer Science Foundation for Research and Technology - Hellas Heraklion, Crete, Greece ************************************************************** =================================== GENERAL =================================== This is sparseLM, a copylefted C/C++ software package for arbitrarily sparse non-linear least squares. IT implements a variant of the Levenberg-Marquardt optimization algorithm that can efficiently deal with non-linear least squares problems giving rise to high-dimensional Jacobians with arbitrary sparseness. Jacobians can be encoded either in compressed row storage (CRS) or compressed column storage (CCS) format. sparseLM can be downloaded from http://www.ics.forth.gr/~lourakis/sparseLM For solving the augmented normal equations arising in the course of the LM algorithm, sparseLM includes interfaces to a wide variety of direct sparse solvers, the list of which currently consists of HSL's MA57, MA77, MA47 & MA27, PARDISO, SuperLU, TAUCS, UMFPACK, CHOLMOD, LDL, CSparse, SPOOLES and MUMPS. While no direct sparse solver is universally superior in terms of performance, the open-source sparse Cholesky factorization package CHOLMOD is reasonably efficient in practice and is hence chosen as sparseLM's default solver. PARDISO (which is integrated into Intel's MKL), is another good commercial alternative. A high-level overview regarding sparseLM can be found in M.I.A. Lourakis, "Sparse Non-linear Least Squares Optimization for Geometric Vision", Proceedings of European Conference on Computer Vision, vol. 2, pages 43-56, 2010. More detailed descriptions will follow in future publications. In case that you use sparseLM in your published work, you are requested to include a reference to the above article: @inproceedings{lourakis10, author = {Manolis I.A. Lourakis}, title = {Sparse Non-linear Least Squares Optimization for Geometric Vision}, booktitle = {European Conference on Computer Vision}, volume = {2}, year = {2010}, pages = {43-56}, doi = {http://dx.doi.org/10.1007/978-3-642-15552-9_4}, } ==================================== FILES ==================================== splm.c: core sparse Levenberg-Marquardt implementation splm_ccsm.c: CCS sparse matrix manipulation routines splm_crsm.c: CRS sparse matrix manipulation routines splm_hessian.c: approximate sparse Hessian J^t*J computation splm_stm.c: Sparse triplet matrix manipulation routines splm_time.c: time measurement utility functions splm_misc.c: miscellaneous functions splm.h: Function prototypes & related data structures splm_priv.h: Private prototypes and definitions for sparseLM splm_cholmod.c: interface to CHOLMOD splm_ldlp.c: interface to LDL splm_csparse.c: interface to CSparse splm_ma27.c: interface to MA27 splm_ma47.c: interface to MA47 splm_ma57.c: interface to MA57 splm_mumps.c: interface to MUMPS splm_pardiso.c:interface to PARDISO splm_dss.c: interface to MKL's DSS interface to PARDISO splm_spooles.c: interface to SPOOLES splm_superlu.c: interface to SuperLU splm_taucs.c: interface to TAUCS splm_umfpack.c: interface to UMFPACK ma77wrap.f90 splm_ma77.c: interface to MA77 splmdemo.c: demo program with examples of sparseLM use matlab/*: Matlab MEX-file interface and demo ============================= COMPILING & LINKING ============================= sparseLM has been developed under Linux and Cygwin with gcc, therefore most of the instructions that follow apply to these environments. Compiling on any other Unix-like system should be straightforward. The software also compiles under MS Windows, however its deployment in this case becomes more complicated due to the fact that most of the required direct solvers do not support compilation in such a setting. Before compiling sparseLM itself, a number of other packages on which it is dependent upon should be compiled. More details are provided below: - First, the desired direct sparse solvers should be specified in solvers.mk. If not already installed, the specified solvers should be compiled following the appropriate steps for each, as described in their documentation. As mentioned above, the suggested solver is CHOLMOD which is part of the SuiteSparse bundle that can be obtained from http://www.cise.ufl.edu/research/sparse/SuiteSparse Unfortunately, the official distribution of SuiteSparse supports compilation only on Linux/Cygwin/Unix systems, not including any support for MS Windows. Nevertheless, compilation instructions for SuiteSparse with MS Visual Studio based on a workaround using Cygwin are at http://matrixprogramming.com/2008/05/umfpack-vc This site has also instructions for compiling MUMPS and TAUCS on Windows. Another option is to download CHOLMOD precompiled for WinXP from http://sites.google.com/site/kevinkaixu/publications/fast/code/cholmod.zip - Any additional packages required by the chosen solver(s) should be installed as well. For instance, both CHOLMOD and HSL's MA57 solvers require METIS from http://www.cs.umn.edu/~metis Please refer to each solver's documentation for more details on required additional software. - LAPACK and BLAS (or equivalent vendor libraries such as MKL) should also be installed since they are needed by all solvers. If not already available, they can be obtained from http://www.netlib.org/clapack and http://www.netlib.org/cblas, respectively. - sparseLM includes makefiles for GNU make (i.e., Makefile/Makefile_demo) that can be used on Linux/Cygwin/Unix and MS nmake (i.e., Makefile.vc/Makefile_demo.vc) that can be used with MS Visual C under Windows. The compilation instructions that follow should be applied on the makefile that is appropriate for your target system. Alternatively, the included CMake configuration file can be used to generate suitable build files; more details can be found below. - After the installation of the sparse solvers, the include paths for their header files should be specified in the configuration section at the top of Makefile. Note that only the paths corresponding to the chosen solver(s) are relevant here; make variables corresponding to unused solvers can point to any path. - In order to link your code against sparseLM, the link paths to the libraries corresponding to the chosen solver(s) and any additional packages should be specified. An example of doing this is provided in the configuration section of the demo program in Makefile_demo. - Finally, typing "make" ("nmake /f Makefile.vc" on Windows) will build both sparseLM and the demo program. - sparseLM can also be built using the CMake cross-platform build system. The included CMakeLists.txt file can be used to generate makefiles for Unix systems or project files for Windows systems that build both sparseLM and the demo program. CMakeLists.txt defines several configuration variables that control which sparse direct solvers are available and the paths where they have been installed. These variables can be modified from CMake's user interface. For example, to compile support for the CHOLMOD linear solver, variable HAVE_CHOLMOD should be set to TRUE and CHOLMOD_INCDIR should be made to point to the directory where CHOLMOD's header files are installed. Defining these two variables suffices to compile the sparseLM library, however linking against it requires two additional variables to be defined: CHOLMOD_LIBDIR, which should contain the path to CHOLMOD's libraries and CHOLMOD_LIBS which should contain their actual names. Support for any other linear solver can be compiled in a similar manner. More informati