Spatio-Temporal Auto Labelling System

We introduce a point cloud labelling system that generates spatio-temporal stability labels by analyzing point-to-point distance correspondence over time. This system requires at least two observations of the same environment taken at different times. The map labelling pipeline is illustrated in the figure below:

Dependencies

To streamline the process, the system should ideally be packaged in a container. Meanwhile, follow these steps to install the necessary dependencies:

Mapping System: This system is compatible with any system capable of generating point cloud maps. In this work, FAST-LIO was used. To set it up, follow these instructions:

cd <your_cw>/src
git clone https://github.com/Livox-SDK/livox_ros_driver.git
git clone https://github.com/ibrahimhroob/FAST_LIO.git
cd FAST_LIO
git submodule update --init
cd ../..
catkin build

Required Libraries: The system requires the OctoMap and PCL (>=1.8) libraries.

sudo apt install ros-$ROS_DISTRO-pcl-*
sudo apt install ros-$ROS_DISTRO-octomap-*
sudo apt install ros-$ROS_DISTRO-octovis   # Useful tool to visualize octomaps

CloudCompare: This tool is highly useful for visualizing point cloud data. It is also used for segmenting the ground plane using CSF. Ensure you have CloudCompare version >= 2.12 installed.

Setup

To build the labelling system, simply use the provided build script:

git clone https://github.com/ibrahimhroob/auto_labeling.git
cd auto_labeling
bash build.sh

Usage

The system accepts rosbags of the environment located in the bags directory within your dataset folder:

DATASET/
├── bags
│   ├── 0.bag
│   ├── 1.bag

To streamline the entire process, we provide a single bash script that includes all the steps for the labelling pipeline along with the necessary hyperparameters. You can find this script in the pipeline.bash file inside the scripts folder. The most important parameters to set are DATASET_DIR and FASTLIO_CW_PTH. Please update these parameters based on your dataset and catkin_ws directories. Here are the main steps in sequence, which you can comment or uncomment based on your testing and requirements:

# Main code
bash create_maps.bash -D $DATASET_DIR \
                      -C $FASTLIO_CW_PTH \
                      -o $OCTO_RESOLUTION \
                      -r $ROSBAG_PLAY_RATE \
                      -m $MAPPING_FILTER_SIZE_MAP \
                      -s $MAPPING_FILTER_SIZE_SURF \
                      -p $MAPPING_FILTER_POINTS \
                      -f $LIDAR_CFG_FILE \
                      -I $IMU_TOPIC \
                      -P $POINTS_TOPIC

# bash extract_scans.bash -D $DATASET_DIR

# bash filter_ground.bash -D $DATASET_DIR \
#                         -r $CLOTH_RESOLUTION \
#                         -t $CLASS_THRESHOLD

# bash tum_to_matrices.bash -D $DATASET_DIR

# bash registration.bash -D $DATASET_DIR -l $DOWN_SAMPLE_FILTER_LEAF_SIZE=0.1

# bash features.bash -D $DATASET_DIR -o $OCTO_DEPTH_QUERY -f $FEATURES_TYPE

# bash label_scans.bash -D $DATASET_DIR -k $KNN_DIS_SCAN_TO_MAP

Publication

If you utilize our code in your academic work, please cite the following paper:

@inproceedings{hroob2023ias,
  author = {Hroob, I and Molina, S and Polvara, R and Cielniak, G and Hanheide, M},
  title = {{LTS-NET: End-to-End Unsupervised Learning of Long-Term 3D Stable Objects}},
  journal = {Proc. of Int. Conf. on Intelligent Autonomous Systems (IAS)},
  year = {2023},
}

@inproceedings{hroob2023ecmr,
  author    = {Hroob, I and Molina, S and Polvara, R and Cielniak, G and Hanheide, M},
  title     = {Learned Long-Term Stability Scan Filtering for Robust Robot Localisation in Continuously Changing Environments}, 
  booktitle = {Proc. of the Europ. Conf. on Mobile Robotics (ECMR)}, 
  year      = {2023},
}