I never thought that I would be writing software to detect guns and weapons from...

https://jamesburnett.io/wp-content/uploads/2020/07/000-gun-detection-tech.mp3

I never thought that I would be writing software to detect guns and weapons from live video feeds. Keeping my family safe at home and in our community has been a top priority. Another priority was to help enhance school security.


After the Stoneman Doublas school shooting I felt I needed to step in and help out in some way. I took some code from a facial recognition project I had worked on and modified it to help detect guns, and it worked. Once I realized that the technology might work, I started a project to prove that it really does work.


The Project.

Several companies offer technology that can detect guns from video feeds. These companies are marketing their software to schools and local governments. Being that I am a steak holder in school security, I wanted to make sure that this technology works before pushing the agenda on to my school board and city council members.


I started off with data collection. Using a replica handgun I took video of myself holding the gun in several different positions. Afterwords I used a tool called image lab to identify the gun in each frame within the video. Those frames were then passed into a convolutional neural network model which took nearly 5 days to compile. I upgraded my GPU to an Nvideo RTX 2060 and the training dropped from 5 days to 4 hours. About 8 months later I ended up with a model that could detect guns from live video feeds.


Imagine for a moment being able to send an alert to a school resource officer or police department. The alert could contain a clip of the object in question for verification and then if necessary, a code red could be issued, potentially stopping a shooting before it could start.


The technology works, I proved that, now it’s just a matter of talking my local government into adopting it.


You can download my gun detection model at my RealTime-Gun-Detection-Software repository.


Video Demo.