project SafeHaven: AI-Driven Offline Security for Every Home

Very cool, ​@shashibhat! I like this a lot!

I run a local repair cafe, and myself and a few of the other tech nerds have a “Think Tank” Signal group where we post about our own tech issues and help each other out 😄 One problem we’ve been trying to solve for a long time, and haven’t successfully done so, is a sensor that shows you if your door is actually locked.

Not just closed, and not just that the handle has been lifted to put the deadbolts in place, but locked. The difficulty is that the lock cylinder has very few moving parts that can be used as an actuator, and those that it does have are inside the door where there’s very little room.

But an AI that monitors the door in a more passive manner could be exactly the solution for this simple, yet seriously confounding problem! 😃 I eagerly await your project!

Hi ​@Spanner ,

I have two kids, so I know exactly what you mean. I actually worked on a similar problem a few years ago. I tried a mechanical approach with a reed switch, but the circuit became way too bulky to fit inside the door frame. And even though I considered using cloud-based AI to detect deadbolt orientation or micro-gaps when the door isn’t fully closed, I didn’t want any camera sending images of my kids to cloud parsers.

With the M.2, I don’t have to worry about that anymore. I can use a bunch of inexpensive cameras near the main door, garage door, and backyard door, and fine-tune the detection using RL — all offline and fully private.

My goal is not to build another “dumb camera alert” system like most security cams. I want it to actually understand why the door is open — whether it’s for a package delivery, kids playing outside, someone doing chores — and only alert when it truly matters. In the future, it could even integrate with smart locks for people who want deeper automation.
 

I can’t wait to get my hands on the kit and start working on this!

Regards,
Shashi

Update:

i just got the kit on new year eve.  So excited and now the fun begins.  
 

Damn, I’m jealous of your kit, and I work here 😄

1. Update:
   With adreline high i started to setup the orangepi with metis, lost remaining few hairs from my bald head figuring out why the orange pi flashing blue light and never turns on. tried everything from changing low res display to lan cable to flashing sd card 101th  time ( just for luck) . 
   later realized it was due to my damn apple power brick. it was not sending PD 5v consistently .
   P.S I am so impressed how thoughtful the organizers are when i saw 5v plug in the kit. with correct power booted up just fine
   
   Given i am working on this after putting kids to bed , i decided to divide the project into two halves
   Hardware stuff:
   I was able to install SDK and about to run the yolo. 
   Since the cameras gonna be on wifi , i found cool tool “angry ip scanner” which give me good way to identify which IPs i need to connect to for cam health checks. working on creating a integrated plugin so that can run lean on the Pi
   
   Software:
   the whole project can only be worth if it has an app.
   1. Monitor, view alerts
   2. train model with custom images so that it can be customized for those edge cases
   3. integration with anything( smart home, triggers or notification type)
   was able to vibe code part of it .now i need come up with a solution of integrating with multiple cams and  find out the break point of yolo or do i need to go to any other . integrating the cam scanning plugin so that can be selected from app itself for the feed. 
   
   Repo: https://github.com/shashibhat/SafeHaven
    

   

   
   Regards,
   Shashi