Inspiration

https://youtu.be/rMQ2b6ZwwCU

In 1986 after lobbying by the National Rifle Association (NRA), the United States Congress passed the Firearm Owners Protection Act (FOPA) which made it illegal for any federal agency to create a centralized database of firearm registrations. When a firearm is bought, the transaction is recorded by hand or type text via form 4473. This form is sent to the Bureau of Alcohol, Tobacco, and Firearms and Explosives (ATF) where it is photographed stored as a JPEG, a non-searchable format.

When a gun crime is committed and a weapon is found at the scene, the ATF agents manually scan through images of form 4473 to find a matching serial number. This labor intensive process requires hundreds of agents and takes weeks to find a single form. Criminals that commit gun crimes are statistically more likely to commit another gun crime shortly after, thus every minute counts when identifying the owner of a gun. Currently, the ATF has over 700 million 4473 gun registration forms. A single gun trace can take anywhere from 3-5 weeks. Each trace request are linked to an active investigation.

What it does

Technology uses computer vision to quickly analyze the contents of the Form 4473 JPEG images and identify a match to any of the ~8,000 active trace request. The application communicates to a mechanical controller via bluetooth to request the next form after analyzing it.

Multiple devices can work in unison to effectively analyze millions of files and reduce a multi-week trace to days or hours. This approach could improve the overall success rate of a gun trace by 4%-12%.

Impact

Prior to entering the TensorFlow Hackathon, we reached out to Giffords law group. They did pro-bono research and determined that our approach does not violate FOPA since we never store any data. Gifford's David Chipman said "This is a truly novel approach that has the potential of saving lives.". We are in direct communication with the ATF and have scheduled a demo with Agent Houser, director of the ATF's National Trace Center.

With Tensorflow 2.0, we will help the ATF trace weapon, apprehend criminals, and save thousands of lives.

Built With

Share this project:

Updates