Team 9 Final Project from EECS568
This repo references a few other repositories which implement the SLAM algorithms used. Use the following command to clone this repository as well as the dependent submodules
git clone --recursive https://github.com/EdWard680/monoslameval.git
Each dataset should go inside its own folder as a bag. Since these bags are large they are not included in this repository, but a link to download the bag is available inside the README within each dataset folder.
These folders also contain the requisite calibration files for each of the algorithms.
Each sub-repo includes instructions for how to build their code which you should follow to build and/or install their code.
To build this you should just need to run
./build.sh
./build_ros.sh
in the ORB_SLAM2 folder as it says. Note that there seems to be a bug on some systems where usleep is not defined, the fix to this is to add
#include <unistd.h>
to the top of include/System.h in the ORB_SLAM2 directory.
Note that lsd_slam depends on very outdated versions of Ubuntu and ROS, and can't really run on the same system as ORB_SLAM2.
Install system dependencies:
sudo apt-get install ros-fuerte-libg2o liblapack-dev libblas-dev freeglut3-dev libqglviewer-qt4-dev libsuitesparse-dev libx11-dev
Find your ROS package path, copy the 'lsd_slam' folder to your ROS package path.
Compile the two package by typing:
rosmake lsd_slam
Not using catkin, however fortunately old-fashioned CMake-builds are still possible with ROS indigo. For this you need to create a rosbuild workspace (if you don't have one yet), using:
sudo apt-get install python-rosinstall
mkdir ~/rosbuild_ws
cd ~/rosbuild_ws
rosws init . /opt/ros/indigo
mkdir package_dir
rosws set ~/rosbuild_ws/package_dir -t .
echo "source ~/rosbuild_ws/setup.bash" >> ~/.bashrc
bash
cd package_dir
Install system dependencies:
sudo apt-get install ros-indigo-libg2o ros-indigo-cv-bridge liblapack-dev libblas-dev freeglut3-dev libqglviewer-dev libsuitesparse-dev libx11-dev
Find your ROS package path, copy the 'lsd_slam' folder under your ROS package path.
Compile the two package by typing:
rosmake lsd_slam
BUG fixing (if error occurs):
If the message like this shows when running command rosmake lsd_slam: 'ERROR [gendeps] 1 Finding dependencies for /home/<yoursettings>/rosbuild_ws/package_dir/lsd_slam/lsd_slam_viewer/cfg/LSDSLAMViewerParams.cfg', then one may fix the problem according to the following instructions:
(1) Open lsd_slam_viewer/cfg/LSDSLAMViewerParams.cfg：
Line 20: scaledDepthVarTH： delete the "'" of the words "point's" and "keyframe's";
Line 21: absDepthVarTH：delete the "'" of the word "point's";
Line 24: cutFirstNKf：delete the "'" of the word "keyframe's".
(2) Open lsd_slam_core/cfg/LSDDebugParams.cfg：
Line 11: plotStereoImages： delete the "'" of the word "what's";
Line 12: plotTracking：delete the "'" of the word "what's";
Line 48: continuousPCOutput：delete the "'" of the word "Keyframe's".
We attempted to set up launch files that could run the datasets for both algorithms, however roslaunch inexplicably doesn't want to start the lsd_slam nodes on ROS indigo, so a separate shell script for starting up the lsd_slam stack using rosrun is included.
To test an algorithm against a dataset, simply run
roslaunch test_dataset.launch dataset:=<dataset folder> algorithm:=<algorithm launch file>
This should work out of the box for the datasets we produced with the M-Bot, however to run on another dataset, you can specify dataset_image_topic to set the image topic which is produced by your bag file. If the topic is of the raw image already, also set decompress:=false.
This will launch rqt_bag on the bag file. Right-click the camera topic (/camera/compressed for the M-Bot datasets) and check the box that says 'publish'. Then you can hit play and you should see the image appear in the viewer window for the algorithm you're running.
Note for ORB_SLAM2, for some reason the window doesn't resize properly. Just close the image viewer window on ORB_SLAM2 and it will re-open
1. Everytime you need to run LSD-slam with a new window, make sure you correctly enter in the terminal:
source ~/rosbuild_ws/setup.bash
sudo chmod u+x lsd_slam.sh
./lsd_slam.sh
Everytime you want to change to run a different dataset, make sure the calibration file 'camera.cfg' is correctly selected.
Reference link: https://github.com/tum-vision/lsd_slam#313-camera-calibration
A useful tool called rpg_trajectory_evaluation is included with this repository. This tool is useful for comparing estimated and ground truth trajectories for the purposes of plotting.
To get groundtruth trajectories from our M-Bot datasets, use the bag_to_pose.py script included in this package. The topic you'll want to convert is the slam_pose topic.
After running ORB_SLAM2 it will save the estimated trajectory to a weird location on your computer. To copy it here, run the provided script
./get_most_recent_orbslam_trajectory.sh
To analyze it, move it into the dataset folder you just ran naming it stamepd_traj_estimate.yaml. Now, you can run the evaluation tool on that folder like so
python2 rpg_trajectory_evaluation/scripts/analyze_trajectory_single.py <dataset folder>
This will generate some analyses and plots in that dataset folder which you can view. Under the hood, this package is running some algorithms to fit the scale of the monocular SLAM trajectory to the ground-truth scale.
See the folder: lsd_plotter for instructions on extracting and comparing the data for lsd_slam
See the lcm_to_ros folder for details about this