leegence/MFG

Multi-Layered Feature Graph for Robot Navigation

MFG

Visual SLAM using Multilayer Feature Graph (MFG)

This is the source code to go along with the IEEE Transactions on Robotics (T-RO) paper by Yan Lu and Dezhen Song, titled "Visual Navigation Using Heterogeneous Landmarks and Unsupervised Geometric Constraints" and published at http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=7103351.

Full citation:

Yan Lu and Dezhen Song, "Visual Navigation Using Heterogeneous Landmarks and Unsupervised Geometric Constraints", IEEE Transactions on Robotics (T-RO), vol. 31, no. 3, June 2015, pp. 736 – 749

This code has been tested on Ubuntu 12.04 and should work on 12.04 or newer versions.

Dependencies

Below is a list of all dependencies required, and links to relevant information such as downloads. There are also instructions for installing the required packages on a Ubuntu system, specifically, but which should also work on most Debian-based systems.

• Qt5: see https://wiki.qt.io/Install_Qt_5_on_Ubuntu
• OpenGL: freeglut (http://freeglut.sourceforge.net/)

  sudo apt-get install freeglut3-dev
  

• OpenCV 2.4.x: http://opencv.org
  • On Ubuntu-based systems, OpenCV should be installed manually from source.
• Line Segment Detector
  • Website link: http://www.ipol.im/pub/art/2012/gjmr-lsd/
  • A copy of the LSD project (version 1.5) resides in this repository and will build with MFG, and includes a custom CMakeLists.txt in order to compile with our code and link correctly
  • Version 1.5 is used for the current state of this project
• LevMar, an implementation of Levenberg-Marquardt Nonlinear Least Squares Algorithms
  • Website link: http://users.ics.forth.gr/~lourakis/levmar/
  • Requires lapack, f2c and OpenBlas

  sudo apt-get install liblapack-dev f2c libopenblas-base libopenblas-dev
  

  • Note that some issues have arisen during the LevMar compilation regarding an undefined reference to symbol 'exp@@GLIBC...', which can be fixed by adding -lm to the TARGET_LINK_LIBRARIES line in the levmar CMakeLists.txt file, making it look like:

  TARGET_LINK_LIBRARIES(lmdemo ${LIBS} -lm)
  

  • Also note that levmar-2.6 has been included and will build with MFG, and contains a slightly modified CMakeLists.txt from the original in order to compile with our code and link correctly
• Eigen3: http://eigen.tuxfamily.org/

  sudo apt-get install libeigen3-dev
  

• g2o: A General Framework for Graph Optimization
  • GitHub download link: https://github.com/RainerKuemmerle/g2o
  • Note that the github code includes CSparse (http://people.sc.fsu.edu/~jburkardt/c_src/csparse/csparse.html)
  • On a UNIX/Linux system:

  git clone https://github.com/RainerKuemmerle/g2o
  cd g2o
  mkdir build
  cd build
  cmake ..
  make
  make install
  

Ensure all installation paths are defined appropriately in the cmake_modules directory's Find[Library].cmake files, if they were installed manually. If the library's path cannot be found, add a line to that library's Find[Library].cmake file for your specific installation path, but please do not commit it to the repository unless it is a generic path (is not under your home directory or in an obscure location).

Build Instructions

This project uses CMake (http://www.cmake.org), a cross-platform build system.

• mkdir build
• cd build
• cmake ..
• make

To run it

1. Under the 'bin' directory, run command: mfg-nogui xxxx where xxxx is a configuration file under 'config'. The default for xxxx is "mfgSettings.ini".
2. To visualize the result, use "src/matlab/plot_mfg_traj.m".