Hello everyone!
So, I recently simultaneously had to upgrade my neural camera detection application for the newest version of OpenVINO, as well as migrate to a fully addon-based setup (due to moving away from running Home Assistant within Proxmox).
This resulted in finally building out my OpenVINO based neural app as a standalone addon, so the supervisor would stop complaining that I’m running an unsupported system.
So, my addon is up and live, and should be usable by anyone running Home Assistant on an amd64 architecture. Since it uses OpenVINO, and OpenVINO is Intel, CPU based neural inference can only work on Intel CPUs. I’m running Home Assistant supervised on an Intel NUC, for example.
If you have an Intel Neural Compute Stick 2, then it might be possible to run the Docker container on a Raspberry Pi as well, but I have not built an image for it yet.
Here’s a link to the repository:
As well as my Home Assistant addons repository:
I’ve tried to make this as standalone as possible. It doesn’t record videos itself, or store image snapshots, etc.
All it does is publish MQTT topics for each camera’s confidence and pixel area values. This is to allow it to work with a variety of setups, and even the Motioneye addon. It also serves an MJPEG stream of all cameras, with the detected bounding boxes, for debugging and seeing what is going on under the hood.
The way I’m using this currently, for example, is that I have Motioneye and this addon running, and then a NodeRED automation that triggers motion in Motioneye when a person is detected, and triggers motion back off after no person is detected for some length of time.
