Drone-vs-Bird: Drone Detection Using YOLOv7 with CSRT Tracker (Official implementation)

Drone-vs-Bird: Drone Detection Using YOLOv7 with CSRT Tracker

(Sahaj K. Mistry, Ajeet K. Verma)

[Paper]

Dataset

The dataset was provided by the WOSDETC workshop & challenge @ICASSP2023 committee. To aquire the dataset you may contact the committee.

Usage

Processing Dataset

To process the dataset, First of all, extract the images using the file extract_images.ipynb then split the dataset using split.ipynb

Training & Testing YOLOv7

After processing the dataset, using main_v7.ipynb to clone, train, and test YOLOv7 on the processed dataset.

Detect Drones Using YOLOv7

Once the YOLOv7 is trained, use the detect.py file from cloned yolov7 folder to detect the drones and use --save_text argument to save the bounding boxes in txt file.

Using Tracker

Use tracker_yolo_output.ipynb along with the video data and generated text file containing bounding boxes to get the CSRT Tracker Powered YOLOv7 Output

Citation

@INPROCEEDINGS{10095146,
  author={Mistry, Sahaj K. and Chatterjee, Shreyas and Verma, Ajeet K. and Jakhetiya, Vinit and Subudhi, Badri N. and Jaiswal, Sunil},

  booktitle={ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 

  title={Drone-vs-Bird: Drone Detection Using YOLOv7 with CSRT Tracker}, 

  year={2023},

  keywords={Surveillance;Signal processing;Acoustics;Object tracking;Speech processing;Drones},

  doi={10.1109/ICASSP49357.2023.10095146}}