School Guard

School shootings are on the rise but computer vision has an answer. Using state-of-the-art machine learning we can teach a computer to find guns in schools and alert people to get to safety.

Comment

Inspiration

So far this year there have been 22 school shootings with casualties and 38 people have been injured or killed. However, this should not be the story. Many schools already have security systems with cameras covering most of the inside and perimeter, just not the people to watch every second for the low chance of an armed intruder. This is where machine learning and artificial intelligence come in.

What it does

School Guard uses a state-of-the-art machine learning neural network to scan camera feeds for guns at 20 times per second. If it finds a gun, it alerts security personnel via text message with the location of the intruder so they can quickly respond with a lockdown and correspond with law enforcement.

How we built it

Neural Networks are built by running thousands of images through an algorithm with the location of the object already identified. The neural network is able to learn how to distinguish this object in new images presented to it. This process takes incredible amounts of time labeling, sorting, and training with excessive compute power required. I used Roboflow to manage a dataset of 3,000 images and then trained a neural network called YOLOv5, part of a state-of-the-art family called you only look once neural Network. These networks only need to see the image once to make an accurate and fast prediction as to whether there is a gun and where. The best part is it is easy to deploy to edge computing devices such as browsers, mobile devices, and on-premise servers.

Challenges we ran into

The model training took much longer than expected and because of the early end time on Sunday morning, this meant I had to let the algorithm run overnight and wake up extra early in the morning to fully implement the product.

Accomplishments that we're proud of

We were able to make a working and precise locator of guns for use in schools. This technology could save thousands of lives in the next years if implemented correctly.

What we learned

We learned that machine learning requires intense compute power at the start but can render some incredible and applicable results.

What's next for School Guard

Next, we will work to get School Guard protecting our schools on-premise. This means deploying edge computing infrastructure to schools in and around the DMV to run the models and notification software to protect our schools.

Built With

Share this project:

Updates