Argos: a spacial-temporal pattern detection system for home automation

I wanted to be able to detect things like if someone exited or entered my main door, whether my dog was having his food, if there were birds in my garden drinking from the dog’s water bowl, if someone forgot to shut off the gas stove after using or any common “pattern” of actions in the house. I have 7 RTMP camera all over the house which till now were only recording, not sensing anything intelligent for HA. so i built this project which uses OpenCV and tensorflow ML, does motion detection triggered object detection, and then pattern detection, notifying homeassistant whenever one is detected via MQTT or webhooks. I had seen and tried https://github.com/blakeblackshear/frigate, but it just does object detection and relied on ffmpeg. You can use argos like an object detector too (yes with your custom TF models as well). Everything is highly configurable and tunable for different use cases.

here’s argos in action:

3 Likes

This sounds like it has lots of potential! Haven’t been attempting room presence ever since the ble solution failed for me (room assistant). Motion sensors aren’t practical in my case too.

I have a spare wyze camera with the custom firmware installed, hence the stream can be accessed with RTSP. That should be fine right?

Will try this out when my pi 4 arrives soon.

this thread is for pattern detection.

A robust and reliable room presence solution with nearly no false negatives! this is the room presence solution built on argos

yes any RTSP/RTMP camera, local raspberry pi camera is fine

see the full argos pattern detection architecture in action in a screen cap of the end to end regression test with real videos from my main door as test cases

you may have to pause the video from time to time to understand what is happening in the console, but the logs are pretty friendly and make it easy to see what’s going on beneath the hood…

the test is running the videos at 10 fps for the purpose of the demo video, otherwise they run at 100+ fps easily.

config for this test is here