DynaLO

This repo contains the implementation for our paper: Dynamic Object-aware LiDAR Odometry Aided by Joint Weightings Estimation in Urban Areas. The proposed DynaLO method is released, and the LiDAR vehicle simulator is now also available in the lidar_vehicle_sim/ directory!

Evaluation

Prerequisites

We tested on Ubuntu 64-bit 18.04, ROS Melodic. ROS Installation. The package is tested on Ubuntu 18.04 with ROS Melodic.

1. Ceres Solver

Ceres Solver

Build

Clone repository:

    cd ~/catkin_ws/src
    git clone https://github.com/DarrenWong/code_for_dynaLO.git
    cd ..
    catkin_make
    source ~/catkin_ws/devel/setup.bash

Download test rosbag

Download dynamic vehicle data, this data is modified based on nuScenes Sequence 0171 using our proposed vehicle simulator (see lidar_vehicle_sim/ for the released simulator).

LiDAR Vehicle Simulator

The lidar_vehicle_sim/ directory contains a ROS-based tool that injects synthetic moving vehicles into a static LiDAR point-cloud scene and produces a ground-truth-labelled ROS bag — without any physical sensor.

Key features:

• Simulates hollow bounding-box vehicle models sampled at 10 pts/m
• Handles occlusion by culling background points inside each vehicle's angular footprint
• Labels points in the time field: 10 = dynamic, 5 = static
• Supports two modes: bag mode (augments a pre-recorded static scene) and live mode (subscribes to /velodyne_points in real time)

See lidar_vehicle_sim/README.md for full setup instructions and usage details. If Git LFS is not familiar or convenient, the simulator input bag non_dynamic.bag can also be downloaded directly from Dropbox.

Launch

    roslaunch dynaLO reweight.launch

Acknowledgements

This work is based on F-LOAM and LIO-Mapping. Thanks for their great work!

Citation

If you use this work for your research, you may want to cite

F. Huang, W. Wen, J. Zhang, C. Wang and L. -T. Hsu, "Dynamic Object-aware LiDAR Odometry Aided by Joint Weightings Estimation in Urban Areas," in IEEE Transactions on Intelligent Vehicles, doi: 10.1109/TIV.2023.3338141.

@article{dynaLO2023huang,
  author={Huang, Feng and Wen, Weisong and Zhang, Jiachen and Wang, Chaoqun and Hsu, Li-Ta},
  journal={IEEE Transactions on Intelligent Vehicles},
  title={Dynamic Object-aware LiDAR Odometry Aided by Joint Weightings Estimation in Urban Areas}, 
  year={2023}
}