We see a lot of car/mail thieves in neighborhoods (esp. here in the bay area) who are checking for unlocked cars in driveway for garage door openers (to get house access) or potentially steal catalytic converters from Hybrid vehicles etc. Most of these thieves operate in a group with a getaway car moving along at slow speeds. Given that most homes have Ring cameras (doorbell + external cameras), if we can enable the Ring cameras in a given neighborhood (taking advantage of Amazon Sidewalk) to opt in to provide an early warning system that can potentially alert authorities to take a look.
We need to obviously rule out false positiives to ensure that the mailman, package delivery or newspaper vendors aren’t flagged but I believe we have enough training samples of car thieves and their modus operandi to make the model more robust.
So, therefore using the devices that someone has configured with Amazon sidewalk, we can use deep learning models in the server to hopefully catch these perpetators in the act and put them behind bars. This can be further optimized with LPR, audio models for glass break etc. to customize the scenarios needed based on what local laws allow.