This is the repository for utilizing pNEUMA Vision dataset.

• Dataset Overview
• Toolbox Overview
• Notes on the Dataset

• This dataset is the expansion of pNEUMA dataset by taking account of imagery parts collected from drones.

• Every 10th frame of the drone videos has been processed with following annotations:

  • NEW Features (only in pNEUMA Vision)
    • images of frame
    • vehicle locations (x, y coordinates in image)
    • azimuths (clockwise, x-axis being 0°)
  • Inherited Features (from pNEUMA)
    • vehicle ID
    • vehicle type
    • timestamp

• You can download the dataset from the following links: link

• If you want to contribute to this dataset or found any issues in this repository please contact us at: [email protected]

• If you are using this dataset, please cite this paper in your work :)

@article{kim2023visual,
  title={Visual extensions and anomaly detection in the pNEUMA experiment with a swarm of drones},
  author={Kim, Sohyeong and Anagnostopoulos, Georg and Barmpounakis, Emmanouil and Geroliminis, Nikolas},
  journal={Transportation Research Part C: Emerging Technologies},
  volume={147},
  pages={103966},
  year={2023},
  publisher={Elsevier}
}

• This toolbox is a collection of functions that can be used to process bounding boxes for pNEUMA Vision dataset as per in the paper.

• You can create two types of bounding boxes: rotated and axis-aligned.

• Session defines the time interval of the video. Please note the session number like following.

  • session 1 is equivalent to 08:00 - 08:30
  • session 2 is equivalent to 08:30 - 09:00
  • session 3 is equivalent to 09:00 - 09:30
  • session 4 is equivalent to 09:30 - 10:00
  • session 5 is equivalent to 10:00 - 10:30

• Example usage of the toolbox is given below.

  • Rotated bounding box

    python process_bounding_box.py --base_dir BASE_DIR --drone DRONE_NUM --session SESSION_NUM
    

  • Axis-aligned bounding box

    python process_bounding_box.py --base_dir BASE_DIR --drone DRONE_NUM --session SESSION_NUM --rotate_bbox
    

  • Save bounding boxes information in a csv file

    The csv file containing bounding box information will be saved in the same directory as the annotations.

    python process_bounding_box.py --base_dir BASE_DIR --drone DRONE_NUM --session SESSION_NUM --save_bbox
    

  • Turn-off visualization while processing

    python process_bounding_box.py --base_dir BASE_DIR --drone DRONE_NUM --session SESSION_NUM --dont_show
    

• 20181029_D3_0900_0930

  • Frames #1616 and #1617 are black frames.

• 20181029_D10_0900_0930

  • Frames #401-#403 and #720-#806 are heavily occluded during the stabilization process due to strong winds.