Helicopters-of-DC Object Detection Dataset and Pre-Trained Model by helicoptersofdc

Images

5483 images

Explore Dataset

From our 2021 Annual Report

Mission

HelicoptersofDC is dedicated to bringing DC residents the most accurate and up-to-date information on what's overhead and to foster collaboration and education about helicopter identification and mission profiles. 2020's GAO report found that 96.3% of helicopter noise complaints to the MWAA from 2018-2021 were unattributed to any specific operator - this is why we feel it's important to educate the community on helicopter identification in addition to reporting what is happening currently and why.

I'd like to thank our Patreon supporters, their contributions cover server costs and allow me to invest in hardware to bring radio calls to the platform.

Unique Strengths of the CopterSpotter System

Many flight tracking sites like Flight Aware, Plane Finder and Radar Box remove flights at the owner's request. Even ADSBExchange, which doesn't filter any flights, does not receive tracking information from most military helicopters due to the Code of Federal Regulations Title 14, Chapter I, Subchapter F rule 91.225 which makes an exemption to mandatory flight transponders for "sensitive government mission for national defense, homeland security, intelligence or law enforcement."

This makes an enormous amount of helicopter flights over the NCR untraceable in any systematic way. As a result, we rely on user submitted photos and identification of helicopters in addition to a separate system that allows users to categorize helicopter radio calls to/from DCA tower.

This dataset is sourced from user-submitted photos on Twitter, and powers a computer vision program to automatically infer submissions to @HelicoptersofDC so DC residents can better identify what's over their heads.

Helicopters of DC (@HelicoptersofDC)

.@fergindc spotted an H60 and has 1997 points 🔭🎷 #CopterSpotter Probable operators: Army/National Guard/FBI/Possibly Presidential twitter.com/a/status/15570… pic.twitter.com/YOmY5VTxoX

Tweet link

Operators around the NCR can be reasonably inferred by aircraft type:

Air Force - UH1(N/H)
Army/Navy/Nat'l Guard - UH60 / H60 / UH72(A) / SH60(m)
Coast Guard - MH65 / AS65
POTUS - V22 / VH60(N) / VH3D / VH92(A)
Medical - EC35 / EC45 / A109(e) / EC135 / EC145
DC/MD/VA Police - AS350 / AS50 / MD52 / A139 / B429
Park Police - B412 / B206

We have identified the following exceptions:

AS50/AS350 is used both by DC Police and at least 3 different DHS helicopters that frequent the area
B412 type is included in Federal Law Enforcement and Emergency support, most frequently Park Police but can also be Dept of Energy's 'Nuke Sniffer' helicopter
B06 is included in Federal Law Enforcement here as Park Police #N33PP. However, it can also be @Chopper4Brad's news helicopter.
B429 included in State and Local for Fairfax County Police, similarly H60 used in Military, but both crafts are used less often by FBI

Future considerations:

This dataset includes classes for three variations of UH60 Black Hawk: H60, VH60N, and H60 + Aux Fuel Tanks. We may merge this later, the initial thought process was the aux fuel tanks change the shape, and the distinct color of the VH60N (when photographed with front light) can be attributed uniquely to Marine Presidential Squadron unlike the rest which can be Army, National Guard, FBI or Customs and Border Patrol.

Release Notes:

August 19th, 2022 1.3 - Adjusted rotation to +/- 5º, retrained model from scratch
August 5th, 2022 1.2 - Now 12,000+ images, added many photos of previously under-represented classes R66, VH92A, S76
Added CH47 class, added bird class, added plane class, added null examples
Rotation: Between -10° and +10°
May 6th, 2022 1.1 - Doubled images to 4061, specifically adding previously under-represented classes such as R66, V22 and A139 - added image augmentations:
- Flip: Horizontal
- Rotation: Between -2° and +2°
- Saturation: Between -50% and +50%
- Exposure: Between -25% and +25%
Trained on Accuracy as opposed to Speed: https://blog.roboflow.com/computer-vision-model-tradeoff/
1.0 - the test version comprised of over 2000 images (that project is now deprecated)

Trained Model API

This project has a trained model available that you can try in your browser and use to get predictions via our Hosted Inference API and other deployment methods.

Cite This Project

If you use this dataset in a research paper, please cite it using the following BibTeX:

@misc{
                            helicopters-of-dc-ghwuq_dataset,
                            title = { Helicopters-of-DC Dataset },
                            type = { Open Source Dataset },
                            author = { helicoptersofdc },
                            howpublished = { \url{ https://universe.roboflow.com/helicoptersofdc/helicopters-of-dc-ghwuq } },
                            url = { https://universe.roboflow.com/helicoptersofdc/helicopters-of-dc-ghwuq },
                            journal = { Roboflow Universe },
                            publisher = { Roboflow },
                            year = { 2024 },
                            month = { apr },
                            note = { visited on 2024-04-18 },
                            }

Connect Your Model With Program Logic

Find utilities and guides to help you start using the Helicopters-of-DC project in your project.

supervision

Find utilities to work with this project in Python.

Roboflow Learn

Build the knowledge you need to evaluate and deploy your model.