Harpy Data – Harpy Data is a Vehicle Dataset with Detector and Tracking Traffic monitoring. Using multiple algorithms (eg. kalman filter) Harpy Data is able to achieve vehicle tracking even when an occlusion occurs. For the detection part we are using Darknet Yolo trained with our own images dataset of vehicles. The velocity calculation algorithm provides very accurate measurements when compared to GPS and odometer data, as demonstrated below. Finally we are able to achieve the fundamental diagram of a road segment using Drones (UAVs) at KIOS Center of Excellence.

About the Dataset

Harpy Data is a Vehicle Dataset with Detector and Tracking monitoring. Using multiple algorithms (eg. kalman filter) Harpy Data is able to achieve vehicle tracking even when an occlusion occurs. For the detection part we are using Darknet Yolo trained with our own images dataset of vehicles. The velocity calculation algorithm provides very accurate measurements when compared to GPS and odometer data, as demonstrated below. You can scroll down to check out the files and information about the Harpy Dataset.

**New data are also available, Traffic Light Status, which are exported by analyzing extracted data from footages using our Harpy Data system. More information can be found in the TLS DATASET page.**

Click here to download Harpy Tracking Open Source code published on Zenodo

What's included in the Harpy data Example Dataset

Raw Videos

The videos we have taken without any detection process. (can be downloaded only with request)

Detection Videos

The videos that have been processed in order to detect cars, their velocity calculations and their count numbers.

Detection Files

All the detection files for each video. The files include velocities, positions (x,y), direction and frame timestamps for each vehicle displayed in the videos.

Updated Demo Version

This is a preview demo of the project’s updated version with improved algorithms for accuracy as well as performance. As you can see roundabouts are also included in the dataset.

More information and results coming soon!!

Example Dataset Results

Paper and citation

Our paper introducing the dataset and the used methods is published at the IEEE VTC 2020 and available here. A preprint on arXiv.org is available here. To reference the dataset, please cite this publication:

@inproceedings{makrigiorgis2020extracting,
  title={Extracting the fundamental diagram from aerial footage},
  author={Makrigiorgis, Rafael and Kolios, Panayiotis and Timotheou, Stelios and Theocharides, Theocharis and Panayiotou, Christos G},
  booktitle={2020 IEEE 91st Vehicular Technology Conference (VTC2020-Spring)},
  pages={1--5},
  year={2020},
  organization={IEEE}
}