As there are shooting incidents happening, it would be great if it could be prevented or help could reach on time. A VWW based system which can detect person with weapons and trigger an alarm or loud sirens could be helpful. A KWS model which can be triggered upon gunshot sounds can also be added. Of course False Positives and False negatives can be extremely costly here. Just a thought. I am not aware if such systems already exists commercially.
I don’t particularly think you need a tinyML device for that camera processing at the endpoint, as that can easily be done with a beefy system plugged into the wall. Of course, there are good reasons to do it without battery power as tinyML devices are easy to just stick to the wall (literally) and therefore easier and cheaper to deploy.
As for the sounds, now that’s really interesting and intriguing because that requires real time processing of the audio. So in that case, you should check out the awesome video by Audio Analytic from my tinyML class last year.
Data Collection Design for Real World TinyML
Sacha Krstulović (Audio Analytic)
Strong encourage everyone to watch it – it was a great lecture by Sacha!!
Thanks for sharing this! It was really informative and helped me gain a whole new perspective and understanding.
There are commercial gunshot detection systems, and they come in two flavours. There are more tactical systems (for example a vehicle-mounted microphone array) that help identify the direction from which a gunshot is fired.
There are also wide-area deployment products (https://www.shotspotter.com/) but as I understand it, there are issues with both privacy (they record) and the potential for discriminatory deployment (decisions on where they are deployed can, in an attempt to target “high crime” areas, also target certain populations by extension.)
I’m interested in the use of sound detection for security purposes in general. For example, acoustic glass break detectors have replaced ugly window foils. Gunshot detection could also be interesting, especially if it is a privacy-protecting design.