An implementation of Graph-based SLAM using just a sequence of image from a monocular camera. It performs feature-based visual odometry (requires STAM library) and graph optimisation using g2o library.
Watch implementation of the algorithm on an aerial robot (Parrot AR.Drone) here
- OpenCV 2.4.x (non-free version)
- LAPACK
- g2o (included. Follow installation instructions)
- Suite-Sparse
- Remove dependency on PCL (not presently using the library any more)
cd <catkin_workspace>/src/
git clone https://justagist@bitbucket.org/justagist/graph_slam.git
## build g2o library ##
cd graph_slam/EXTERNAL/g2o/
cmake . && make
## clone STAM library (branch - graphslam_mod) for visual odometry ##
cd ..
git clone -b graphslam_mod https://github.com/eaa3/STAM.git
cd STAM/
mkdir build && cd build
cmake ..
make -j
## build package ##
cd <catkin_workspace>/
catkin_make
source devel/setup.bash
Before running the graph-slam node, the location of the 'fabmap_training_data' folder has to be entered by editing the value of 'fabmap_training_data_destination_' parameter in graph_slam/src/main.cpp file. The path to the image dataset on which the algorithm is to be run can also be set in the main.cpp file.
rosrun graph_slam main_slam_node <scene_number> [visualize?] [publish rostopics?] [save trajectory to txt file?] [run graph optimisation thread?] [baseline for visual odometry] [trajectory file name]
Run rosrun graph_slam main_slam_node for detailed usage instructions
- Plot trajectory output from graph_slam:
rosrun graph_slam trajectory_plotter.py <trajectory_file.txt> - Plot and compare 2 trajectories:
rosrun graph_slam compare_trajectories.py <trajectory_file_1.txt> <trajectory_file_2.txt> - Compare trajectory with ground_truth (if available):
rosrun graph_slam groundtruth_compare.py <trajectory_file.txt> <groundtruth.txt>
Function and usage of all nodes are described in the respective source files, along with the format of the input files (where required).