Developer: Wenyu Li, Xinyu Chen, Yantao Yu
Affiliation: The Hong Kong University of Science and Technology
This is a description of our dataset paper: A real world visual slam dataset for indoor construction sites
The dataset will be available after the paper is accepted.
| Sequence | Duration | Size | Description | Link |
|---|---|---|---|---|
| Building A1, static | 85s | 9.30G | Texture-less, multiple rooms | Rosbag |
| Building B1, static | 78s | 8.30G | Repetitive texture, looped trajectory | Rosbag |
| Floor14, static2 | 56s | 7.32G | Over-exposure | Rosbag |
| Floor4, static1 | 35s | 4.56G | Glass, static workers | Rosbag |
| Floor4, static3 | 24s | 3.08G | Texture-less, single room | Rosbag |
| Floor1, dynamic | 55s | 5.43G | Motion blur, dynamic workers | Rosbag |
- The Floor1 sequence is a segment of the raw data and does not contain tf messages. Transformations from other sequences can be used.
The dataset was collected using a RealSense L515 camera and an Ouster OS0-128 LiDAR. Further details about the sensors can be found in our paper.
RealSense L515 | Ouster OS0-128
Sensor Data
The dataset is recorded in rosbag format. We recommend using Ubuntu and ROS for optimal compatibility. The system has been tested on ROS Noetic and Ubuntu 20.04, but other versions should also work.
To use the provided lidar slam ground truth and visual slam benchmark, you need to install docker on your computer. If you are going to run them on your host machine, you might need to do some modification ( follow the original repositories or instructions in our dockerfile).
You need to check if these systems support your ubuntu or ros version. Melodic or Noetic should be a good choice. Make sure your could run popular slams on your computer and just directly use our provided rosbags in your own system.
If you are tired of struggling with the environments required by different repositories, you can have a try on our docker scripts.
Install docker engine and nvidia-docker(optional, but recommend), please follow the official docs. Do NOT install docker desktop.
https://docs.docker.com/engine/install/ubuntu/
https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#docker
After installation, cd to the corresponding docker folder on your host machine, you will see three files. Change the name of docker image, docker container, path folder on your computer, etc.
├── ros_package_xxx/
│ ├── docker/
│ │ ├── build.sh
│ │ ├── Dockerfile
│ │ └── run.sh
│ ├── launch
│ ├── src
│ ├── ...
│ ├── CMakeLists.txt
│ └── package.xml
You can also run on a machine without nvidia display card. Just try :
# delete "--gpus all " in the run .sh file.
docker run -it --name $container_name --gpus all \
# delete these lines in Dockerfile
ENV NVIDIA_VISIBLE_DEVICES ${NVIDIA_VISIBLE_DEVICES:-all}
ENV NVIDIA_DRIVER_CAPABILITIES ${NVIDIA_DRIVER_CAPABILITIES:+$NVIDIA_DRIVER_CAPABILITIES,}graphics
Compile docker image on your host computer. It will take some time, depending on your internet.
bash build.sh
Start a new docker container. After that you can use the container(See docker official docs). You can use run.sh for multiple times. The old container will be instead with the new one.
bash run.sh
In our setting container, /root is your working space. In the /root folder, catkin_ws is your ros workspace; dataset and downloads are connected to your host machine; start.sh is a starting script for compiling and preparation, you can change it at the end of the dockerfile.
To enable rviz in the container, enter on your host machine.
xhost +
If you want better support on the sensors, you may also want to install their ros drivers. To install on your host machine or docker container:
For RealSense:
apt-get update && apt-get -y install apt-transport-https && \
mkdir -p /etc/apt/keyrings && \
curl -sSf https://librealsense.intel.com/Debian/librealsense.pgp | tee /etc/apt/keyrings/librealsense.pgp > /dev/null && \
echo "deb [signed-by=/etc/apt/keyrings/librealsense.pgp] https://librealsense.intel.com/Debian/apt-repo `lsb_release -cs` main" | \
tee /etc/apt/sources.list.d/librealsense.list
apt-get update && apt-get install -y \
librealsense2-dkms \
librealsense2-utils \
librealsense2-dev
apt-get update && apt install -y ros-melodic-realsense2-camera
For Ouster:
apt-get update && apt-get install -y \
ros-noetic-pcl-ros \
ros-noetic-rviz \
libspdlog-dev
Download in your ros workspace.
git clone --recurse-submodules https://github.com/ouster-lidar/ouster-ros.git
You are encouraged to use your own lidar slam as the ground truth based on our provided lidar data for higher accuracy evaluation. But here we also offer you an out of the box package configured in docker. It originates from LIMO VELO, a specific lidar SLAM modified from FAST-LIO2.
In the docker container, modify the launchfile and then run :
roslaunch limovelo ouster.launch
In our paper, we used VINS-RGBD as a baseline. However, we also provide both ORBSLAM3 and VINS-RGBD for benchmarking. They run in RGBD-Inertial mode.
Open the corresponding container and run:
roslaunch vins_estimator l515.launch
or
roslaunch orbslam3 vio.launch
For vins rgbd with rtabmap：
For orbslam3 RGBD-I mode (it failed):
The localization results are saved in TUM format. You can use the evo tool for trajectory evaluation. Note that the extrinsic calibration between the LiDAR and the camera is not provided. For alignment in evo, use the --align option for automated alignment, or manually set a common starting point.
We would like to thank the contributors of the referenced repositories. Please cite our paper if this dataset is useful for your work.
If you have any problems with the dataset or software, please open an issue.
For academic or commercial collaboration, please contact the corresponding author Prof. Yantao Yu.