kminoda / VIODE

VIODE: A Simulated Dataset to Address the Challenges of Visual-Inertial Odometry in Dynamic Environments

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

The VIODE dataset

This is a repository for the VIODE (Visual-Inertial Odometry in Dynamic Environments) dataset described in the paper:

Koji Minoda, Fabian Schilling, Valentin Wüest, Dario Floreano, and Takehisa Yairi, VIODE: A Simulated Dataset to Address the Challenges of Visual-Inertial Odometry in Dynamic Environments, IEEE Robotics and Automation Letters (RA-L), 2021. PDF

The overall documentation is available in the above RA-L paper. If you use VIODE in academic work, please cite:

@article{minodaRAL2021,
  title={{VIODE}: A Simulated Dataset to Address the Challenges of Visual-Inertial Odometry in Dynamic Environments},
  author={Minoda, Koji, and Schilling, Fabian, and W\"{u}est, Valentin, and Floreano, Dario, and Yairi, Takehisa},
  journal={IEEE Robotics and Automation Letters},
  year={2021},
  volume={6},
  number={2},
  pages={1343-1350},
  doi={10.1109/LRA.2021.3058073}}
 }

A YouTube video for an introduction to the VIODE dataset:
VIODE

Dataset Link

Main ROS bag files are uploaded on Zenodo:
Data download here (A link to Zenodo)

The other two files can be downloaded from this repository.

Dataset Structure

The visual-inertial sensor data is provided in ROS bag format. Each bag contains the following topics.

  • /cam0/image_raw
  • /cam1/image_raw
  • /cam0/segmentation
  • /cam1/segmentation
  • /imu0
  • /odometry

/cam0/image_raw and /cam1/image_raw contain RGB image data. Since these are captured in the simulator, we also provide ground-truth extrinsic and intrinsic parameters for this stereo setup.

/cam0/segmentation and /cam1/segmentation are the ground truth semantic segmentation provided by AirSim. The ex-/intrinsic parameters are the same as the ones with the RGB images.

The other files are

  • calibration.yaml
  • cam0_pinhole.yaml & cam1_pinhole.yaml
  • rgb_id.txt
  • vehicle_ids_*.txt

The calibration.yaml, cam0_pinhole.yaml, and cam1_pinhole.yaml provide extrinsic and intrinsic parameters of two cameras. Note that the format follows that of VINS-Fusion, one of the state-of-the-art VIO algorithms. The rgb_id.txt provides the correspondence between object ID and RGB value in segmentation images. The vehicle_ids_*.txt contains the correspondence between object ID and object name for each environment. Object name also indicates whether the vehicle is whether dynamic or static.

About

VIODE: A Simulated Dataset to Address the Challenges of Visual-Inertial Odometry in Dynamic Environments