/drone-tracking-datasets

Multi-view drone tracking datasets

Primary LanguageShellOtherNOASSERTION

Multi-view drone tracking datasets

UPDATE!!! - Manual labels of drone locations in all datasets complete! Enjoy!

This repository contains datasets where a flying drone (hexacopter) is captured with multiple consumer-grade cameras (smartphones, compact cameras, gopro,...) with highly accurate 3D drone trajectory ground truth recorderd by a precise real-time RTK system from Fixposition. In some datasets, the ground truth temporal synchronization and ground truth camera locations are also provided. The table below gives a brief summary of what kind of data is available for each dataset.

Dataset 3D trajectory Synchronization Camera locations 2D labels 3D orientation
1 Yes No No Yes No
2 Yes No No Yes No
3 Yes Yes Yes Yes No
4 Yes Yes No Yes No
5 Yes Yes Yes No Yes

We provide calibration parameters as well as calibration images for each camera.

If you use this dataset in your research, please cite our paper in which we estimate 3D trajectories from multiple unsynchronized cameras.

The pipeline we developed can be found here .

Installation

The videos have been split into multiple archives using zip. To extract, use 7z or other zip archive manager or run the provided script ./unpack.sh from the repository root directory which will extract all videos for you, provided that you have the latest 7z installed.

Dataset 1

Easy dataset with short and slow flight.

# of cameras Flight duration
4 ~ 2min

Camera 0 trajectory Camera 1 trajectory Camera 2 trajectory Camera 3 trajectory

Dataset 2

Easy dataset with longer and faster flight.

# of cameras Flight duration
4 ~ 2.5min

Camera 0 trajectory Camera 1 trajectory Camera 2 trajectory Camera 3 trajectory

Dataset 3

Medium difficulty dataset with more cameras, long flight, various velocities and motion types.

# of cameras Flight duration
6 ~ 9min

Camera 0 trajectory Camera 1 trajectory Camera 2 trajectory Camera 3 trajectory Camera 4 trajectory Camera 5 trajectory

Dataset 4

High difficulty dataset with many cameras, fast motion and moving clouds making the detection harder.

# of cameras Flight duration
7 ~ 7min

Camera 0 trajectory Camera 1 trajectory Camera 2 trajectory Camera 3 trajectory Camera 4 trajectory Camera 5 trajectory Camera 6 trajectory

Dataset 5

A new challenging winter dataset with 6 cameras, snow covered background and multiple drones. The ground truth drone trajectory is estimated by fusing total station tracking and onboard IMU data.

# of cameras # of drones Flight duration
6 3 ~ 10min

Dataset 5 Overview

Please check here for the principle and toolkits used for collecting and processing dataset 5.

Authors

  • Cenek Albl (cenek.albl-at-gmail.com) - project lead
  • Jingtong Li (tong226-at-hotmail.com) - data collection, GNSS, camera calibration
  • Jesse Murray (jesse.r.murray-at-gmail.com) - data collection, camera calibration
  • Chen-Chieh Liao (liao86632769-at-gmail.com), Dorina Ismalii (dorinaismaili01-at-gmail.com ) - 2D trajectories
  • Mudathir Awadaljeed (awadaljm-at-student.ethz.ch) - Ground truth 2D labelling
  • Yue Pan (yuepan-at-student.ethz.ch) - data collection of dataset 5

The work was conducted in the group of Photogrammetry and Remote Sensing, ETH Zurich.

Acknowledgement

Special thanks to

  • Fixposition for providing the drone and the RTK system
  • Alexander Wolf for the synchronization hardware
  • Riccardo De Lutio, Tom Manu, Stefano D'Aronco, Mikhail Usvyatsov for helping with data collection