Over the past couple of months, it’s been pretty chaotic with school, robotics, tennis, and just life. I found some time to work on the project with my dad, who is really good at networking. After figuring out how to install TensorFlow on the Raspberry Pi, we ran a gunshot-detection model I found online. Hooked up to the SiZheng microphone, a total of three sub-models were run.
- 44100 x 1 model
- 128 x 64 model
- 128 x 128 model
To avoid false positives, all three models have to detect a gunshot for it to be true. If only one or two models detect a gunshot, it will be classified as a loud noise and disregarded. I have tested this methodology with recordings of gunshots, clapping hands, and other loud noises around the house. The models can run constantly and have only crashed once so far.
Some learning curves were understanding how much computing power a Pi Zero had versus a Pi 4, and knowing which to purchase. Some upcoming posts will include intense testing of the device and a computational science study I’m working on at NCSSM regarding factors affecting gun violence incidents.
Next, I’ll be creating a dashboard so that school or building administration can view and access live data at all times locally. Setting up a notification loop will also be good, so admin doesn’t have to always be checking.
