RGBD-SLAM

This program is a simple RGBD-SLAM demo. It uses RGBD images (such as Kinect images) as input, performing tracking and mapping simultaneously. I finished and improved this program based on Gao Xiang's tutorial in cnblogs (in Chinese).

This SLAM program uses PnP algorithm in tracking (front end) and global Pose graph optimization in the back end. This program supports ROS (Robot Operating System) and can visualize the result (such as trajectories and dense point cloud map) in rviz. Here is a demo video on youtube.

0. Prerequisites

Ubuntu 64-bit 16.04. ROS Kinetic. ROS Installation

G2O. A modified version of g2o is in Thirdparty directory, compile and install it in your computer.

DBoW3. A famous package for loop closure. And a large binary ORB vocabulary file can be found in vocab directory.

1. Installation

Create your catkin workspace:

mkdir -p ~/catkin_ws/src 
cd ~/catkin_ws/
catkin_make
echo "source ~/catkin_ws/devel/setup.bash">> ~/.bashrc 
source ~/.bashrc

In your ROS package path, clone the repository:

cd ~/catkin_ws/src
git clone https://github.com/gjgjh/rgbdslam-demo

Compile the rgbdslam package by typing:

cd ~/catkin_ws/
catkin_make

2. Usage

Before running the program, you may need to adjust some parameters setting in config/default.yaml. For detailed information, you can look into the comment.

After that, simply launch the launch file in the terminal:

roslaunch rgbdslam rgbdslam.launch

Download the rgbd dataset from TUM website: https://vision.in.tum.de/data/datasets/rgbd-dataset Then, you can play the rgbd dataset by typing these in terminal, for example:

rosbag play YOUR_DATA_FOLDER/rgbd_dataset_freiburg1_desk.bag

You will see two windows showing the current rgb and depth frame, and another window showing the 3D map and trajectories in real time. Before you shutdown the node, you can type these in a new terminal to save the final results:

roslaunch rgbdslam map_saver.launch

The optimized trjectories(.txt), 3D point cloud map(.pcd) and octomap(.ot).You can use pcl_viewer to view the point cloud map, and use octovis to view the octomap. After all data is done and your results are saved, you can press ctrl+C to stop these nodes. If you want to evaluate the SLAM results, there are some tools in the /scripts directory for the benchmark. For example, you can use it by typing (ground truth file can be found on TUM website):

python2 scripts/evaluate_ate.py scripts/rgbd_dataset_freiburg1_desk-groundtruth.txt result/CameraPoses.txt --plot figure.png --offset 0 --scale 1 --verbose

You can also use the online evaluation provided by TUM website. Note that the error metric is independent of the coordinate system, since the true and the estimated trajectory will be aligned beforing evaluating.

3. License

The source code is released under GPLv3 license.

We are still working on improving the code reliability. For any technical issues or commercial inquiries, please contact GJH guojh_rs@pku.edu.cn.