SVIO

A RGB-D Visual-Inertial Odometry Leveraging Structural Regularity

Related Paper

Pengfei Gu and Ziyang Meng, "S-VIO: Exploiting Structural Constraints for RGB-D Visual Inertial Odometry", IEEE Robotics and Automation Letters, 2023. pdf

1. Prerequisites

1.1 Ubuntu and ROS

Test on Ubuntu 18.04 and ROS Melodic. Follow ROS Installation. The following ROS pacakges are needed:

    sudo apt-get install ros-YOUR_DISTRO-cv-bridge ros-YOUR_DISTRO-tf ros-YOUR_DISTRO-message-filters ros-YOUR_DISTRO-image-transport

1.2 Ceres Solver

Follow Ceres Installation.

2. Build SVIO on ROS

Clone the repository and catkin_make:

    cd YOUR_PATH_TO_SVIO/SVIO/src
    git clone https://github.com/oranfire/SVIO.git
    cd ../
    catkin_make

Before running the SVIO, remember to source the script first:

    source YOUR_PATH_TO_SVIO/SVIO/devel/setup.bash

3. Usage

3.1 VCU-RVI handheld dataset

Download VCU-RVI Dataset. We use the time-synchronized IMU measurements and RGB-D images for localization. Open two terminals, launch the estimator and play the bag respectively. Take motion_1 for example:

    roslaunch YOUR_PATH_TO_SVIO/SVIO/src/launch/svio_vcu_run.launch
    rosbag play YOUR_PATH_TO_BAG/lab-motion1.bag

It will automounsly launch the rviz for visualization.

3.2 OpenLORIS-Scene wheeled robot dataset

Download OpenLORIS-Scene Dataset. Although it contains multiple sensors, we only use the RGB-D images and IMU measurements from the d400 depth camera. Open two terminals, launch the estimator and play the bag respectively. Take home1-1 for example:

    roslaunch YOUR_PATH_TO_SVIO/SVIO/src/launch/svio_openloris_run.launch
    rosbag play YOUR_PATH_TO_BAG/home1-1.bag

Note that the bag files provided by the OpenLORIS-Scene dataset record the acceleration and the angular velocity in two topics. Therefore before playing the bag, we should merge these two topics into one topic by using the script provided by OpenLORIS-Scene Tools first:

    python merge_imu_topics.py YOUR_PATH_TO_BAG/home1-1.bag

3.3 EuRoC MAV dataset

As a bonus, SVIO also provides interface for a stereo-inertial dataset, EuRoC Dataset, where the depth image is computed from the rectified stereo images using the SGM algorithm. Take mh01 for example:

    roslaunch YOUR_PATH_TO_SVIO/SVIO/src/launch/svio_euroc_run.launch
    rosbag play YOUR_PATH_TO_BAG/mh01.bag

4. Credits

Many thanks to the authors of VINS-Mono, DUI-VIO and ManhattanSLAM. Our system is built on the first two projects, and part of codes are borrowed from the third project.

5. Citation

    @ARTICLE{10107752,
      author={Gu, Pengfei and Meng, Ziyang},
      journal={IEEE Robotics and Automation Letters}, 
      title={S-VIO: Exploiting Structural Constraints for RGB-D Visual Inertial Odometry}, 
      year={2023},
      volume={8},
      number={6},
      pages={3542-3549},
      doi={10.1109/LRA.2023.3270033}}

6. Licence

The source code is released under GPLv3 license.