#RGBDTAM:
RGBDTAM is a RGBD SLAM algorithm that estimates a dense reconstruction of a scene in real-time on a CPU using monocular or RGB-D cameras. We are currently cleaning the code and working on the efficiency of the algorithm, it should be ready in the following days/weeks.
Related Publication: [1] Alejo Concha, Javier Civera. RGBDTAM: A cost-effective and accurate RGBD Tracking and Mapping System. Submitted to RA-L with IROS presentation.
Video of the results that you should expect in the example sequences: Soon
#License
RGBDTAM is licensed under the GNU General Public License Version 3 (GPLv3), please see http://www.gnu.org/licenses/gpl.html.
For commercial purposes, please contact the authors.
#Disclaimer
This site and the code provided here are under active development. Even though we try to only release working high quality code, this version might still contain some issues. Please use it with caution.
#Dependencies
ROS:
We have tested RGBDTAM in Ubuntu 14.04 with ROS Indigo.
To install ROS (indigo) use the following command:
sudo apt-get install ros-indigo-desktop
Or check the following link if you have any issue:
http://wiki.ros.org/indigo/Installation/Ubuntu
PCL library for visualization:
sudo apt-get install ros-indigo-pcl-ros
BOOST library to launch the different threads:
sudo apt-get install libboost-all-dev
Vocabulary used for loop closure and relocalization:
We have used the vocabulary created by ORB-SLAM authors. Please, download the vocabulary from this link "www.github.com/raulmur/ORBvoc.txt.tar.gz" and place it in "ThirdParty/DBoW2/build/ORBvoc.txt"
#Installation
git clone https://github.com/alejocb/rgbdtam.git
#Compilation
catkin_make --pkg rgbdtam
Third Party: SUPERPIXELS COMPILATION
Code used -> Efficient Graph-Based Image Segmentation. P. Felzenszwalb, D. Huttenlocher. International Journal of Computer Vision, Vol. 59, No. 2, September 2004
cd root/catkin_workspace/src/rgbdtam/ThirdParty/segment
make
#Usage
Launch rgbdtam from your 'catkin_workspace' folder:
cd root/catkin_workspace
rosrun rgbdtam rgbdtam
Notice that the location of rgbdtam should be the following:
root/catkin_workspace/src/rgbdtam
Launch the visualizer of the current frame
rosrun image_view image_view image:=/rgbdtam/camera/image
Launch the visualizer of the map
rosrun rviz rviz
We are working on an automatic visualizer, but for now, check the following screenshot to set up the rviz visualizer:
https://www.dropbox.com/s/pymufqi2i2aixys/visualization_rviz.png?oref=e&n=314995776
You can use a sequence from the TUM dataset to test the algorithm:
rosbag play sequence.bag
There are two parameters that you have to modify (before executing a sequence) in rgbdtam/src/data.yml:
1-) Intrinsic parameters of the camera:
'cameraMatrix'
'distCoeffs'
2-) Camera topic
camera_path:"/image_raw"
Update the the 'camera_path', 'cameraMatrix' and 'distCoeffs' in the file rgbdtam/src/data.yml
Parameters
There are a few tuneable parameters that you can modify in rgbdtam/src/data.yml:
1-) Superpixel calculation
calculate_superpixels: [bool] If 1 it will calculate 3D superpixels.
2-) Number of frames for mapping
num_cameras_mapping_th: [int]. Number of frames that you want to use to estimate the depth maps. Default: 9.
3-) Minimum parallax required for mapping
translational_ratio_th_min: [double]. Minimum parallax to insert a keyframe. Default: 0.075. Typical values [0.03-0.15].
4-) Degenerated cases in 3D superpixel matching
limit_ratio_sing_val: [double]. This threshold deals with the degenerated cases in 3D superpixel calculation. Smaller values -> less outliers. Default: 100. Typical values [10-1000].
5-) Minimum normalized residual threshold required.
limit_normalized_residual: [double]. This threshold accounts for the minimum error required in superpixel calculation. Smaller values -> less outliers. Default: 0.30. Typical values [0.05-0.50].
6-) Minimum number of matches of 3D superpixels in multiple views to achieve multiview consistency.
matchings_active_search: [int]. Number of matches required of the 3D superpixel in multiple views. Larger values -> less outliers. Default: 3. Typical values [0-4].
7-) Kinect Initialization: 1 kinect_initialization: [bool] If 1 it will use the kinect for initialization.
8-) kinect_initialization: 1 minim_points_converged: 66
Contact
If you have any issue compiling/running rgbdtam or you would like to know anything about the code, please contact the authors:
Alejo Concha -> aconchabelenguer@gmail.com
Javier Civera -> jcivera@unizar.es