Locally processed AI for security cameras


I want to run an AI stack locally to perform a person and object detection on multiple camera streams. I know that there is a blue-iris add-on but I don’t use blue-iris and the add-on didn’t work good for my friend.
I would like to have all in one solution(hardware and software) with a nice user interface. Any suggestions?

I see cloud-based solutions but it becomes very expensive over time. I also don’t like to share my video streams to the cloud.

I would like to start a project if there is no product that can do this
This is what I imagine;

  • Support at least six 1080p streams
  • Below 30W consumption, 10W will be really good.
  • Nice web interface
  • SMS, email, push notifications, mqtt etc
  • Linux based and allow custom add-ons for future proof
  • Less than $1000

Please share your experience and thoughts on this topic and let me know if you are interested in participating this project.

I am not aware of any systems like that.

Does blue iris do any of these things?

Or motioneye?

If it helps, my security system for my van consists of 4 phones streaming over my raspberry pi bluetooth network running a basic version of fedora that’s piped to a 128GB flash drive for video cache.

I’m working on having it utilize xfinity wifi hotspots so I can stream footage to my smartphone while in per say a coffee shop with my home on wheels parked around the corner.

I know too many friends who’ve had their vehicles broken in to while in the city which is what drove me to build this extremely low-powered video insurance system

That’s a great solution.


This one is for Blue Iris. Middleware is open-source however DeepStack is closed source. (It’s free to get an API key for now.)

Note: Deep stack is abandoned for a long time and I believe if the dev team ever come back, they will start charging for an API key because of this project :slight_smile:

Here is what I think about this setup;
-It will require a beefy processor to handle 2+ cameras. You will need to have a high-end i7 to run blue iris and this AI tool with DeepStack.
-Because everything is running on a CPU, I believe developers decided to only analyze motion alerts rather than analyzing the real-time stream. If blue iris misses a motion then the tool won’t help you.
-Limited AI functionality. (Who knows, maybe someone will adopt another engine into this tool in the future)
-High power consumption.

Are you running detection software on your phone?

Zoneminder https://zoneminder.com/ has an machine learning addon for this https://github.com/pliablepixels/zmeventnotification

You would just need some hardware to put it on. I think for most of these (excluding blue iris which requires windows ) you would use similar hardware.

No, but I’ve been thinking about building a little A.I. on my desktop running fedora Workstation to recognize faces and porting it to the raspberry pi running silverblue.

The raspberry pi is the main pipe for the rest of the phone cameras, so I could put priority to processing the video stream(s) that have movement.

Maybe a simpler method would be to connect bluetooth motion sensors to the raspberry pi and have it send a notification to my phone when it trips