Skip to content

Latest commit

 

History

History
64 lines (54 loc) · 2.13 KB

File metadata and controls

64 lines (54 loc) · 2.13 KB

Multi-view Drone Tracking Datasets (GitHub)

Overview

This repository contains datasets where flying drones (hexacopters) are captured with multiple consumer-grade cameras (smartphones, compact cameras, GoPro, etc.) with highly accurate 3D drone trajectory ground truth recorded by a precise real-time RTK system from Fixposition.

Dataset Components

The repository contains 5 datasets with varying characteristics:

Dataset 1

  • 4 cameras
  • Flight duration: ~2 minutes
  • Easy dataset with short and slow flight
  • 3D trajectory ground truth: Yes
  • 2D labels: Yes

Dataset 2

  • 4 cameras
  • Flight duration: ~2.5 minutes
  • Easy dataset with longer and faster flight
  • 3D trajectory ground truth: Yes
  • 2D labels: Yes

Dataset 3

  • 6 cameras
  • Flight duration: ~9 minutes
  • Medium difficulty dataset with more cameras, long flight, various velocities and motion types
  • 3D trajectory ground truth: Yes
  • Synchronization: Yes
  • Camera locations: Yes
  • 2D labels: Yes

Dataset 4

  • 7 cameras
  • Flight duration: ~7 minutes
  • High difficulty dataset with many cameras, fast motion and moving clouds making detection harder
  • 3D trajectory ground truth: Yes
  • Synchronization: Yes
  • 2D labels: Yes

Dataset 5

  • 6 cameras
  • 3 drones
  • Flight duration: ~10 minutes
  • Challenging winter dataset with snow-covered background and multiple drones
  • 3D trajectory ground truth: Yes
  • Synchronization: Yes
  • Camera locations: Yes
  • 3D orientation: Yes

Technical Details

  • Calibration parameters and calibration images provided for each camera
  • Manual labels of drone locations in all datasets
  • Videos split into multiple archives using zip format
  • Ground truth drone trajectory in dataset 5 estimated by fusing total station tracking and onboard IMU data

Characteristics for Drone Deployment

  • Multiple camera views provide diverse perspectives for tracking
  • Varying difficulty levels from easy to challenging scenarios
  • Different environmental conditions (including winter/snow)
  • Multiple drones in dataset 5 for multi-object tracking
  • Precise ground truth for algorithm evaluation

Source

GitHub: https://github.com/CenekAlbl/drone-tracking-datasets