Sean Ferigan: Home Automation

 Not much progress going on with my home automation recently. Only thing of note is that I've migrated most of my Amazon smart plugs over to Innr Zigbee. This removes more devices from my 2.4g wifi network, plus the Amazon smart plugs never did seem to get a great signal even though I've got a rock solid UniFi wifi network around the house.

I've not been able to spend too much time on my home automation the last couple of weeks due to work commitments but have just sat down for 30 minutes and added a much needed feature to my lighting setup. Prior to now, I've had most of my lighting triggered on motion all controlled through Node Red flows. This works really well and, depending on time of day, it selects different moods (I use Philips Hue bulbs and sensors but have the motion sensors connected to a Conbee II to get an almost instant reaction time compared to using the Hue API). Anyway, back to the feature. I'll use my office as an example. Most of the time it's using subdued lighting to give effect rather than high light levels. This is great when I'm working or gaming but sometimes I just need the lights to be bright (a recent example was when transplanting my main PC into a new case). My solution is to have a boolean toggle to override the lighting to a separate "bright" mood. The motion d...

 Just a quick update on my progress migrating from MotionEye to Frigate for my IP camera motion detection / camera feeds. Setting up Frigate was super easy. I had it processing the camera feeds and detecting people in just a few minutes offloading all of the heavy ML pieces to the Google Coral USB. The bit that I've just spent a couple of hours toying with is changing my Home Assistant Lovelace dashboards over to using the Frigate camera feeds. Setting it all up was easy, installed the Frigate HA integration which created all of the HA devices etc, updated my Lovelace cards to show the camera feeds and it all appeared to work. The bit I'm pulling my hair out over at the moment is that the live camera feeds really lag on my phone and PC (phone connected via WiFi, PC on a wired LAN connection). My gut feel atm is that it's because of the image sizes (2 of my cameras are 2k resolution) so I'm going to see if there's a way to downscale them for the dashboards. I'll ...

 My Google Coral Edge TPU arrived today which I'm hoping will make person detection on my CCTV fast and simple, and reduce the number of false positives I get currently. Here's a quick unboxing video. And if you're thinking to yourself "What the heck is that??" then here's a little blurb: The Coral USB Accelerator adds an Edge TPU coprocessor to your system. It includes a USB socket you can connect to a host computer to perform accelerated ML inferencing. The on-board Edge TPU is a small ASIC designed by Google that provides high performance ML inferencing with a low power cost. For example, it can execute state-of-the-art mobile vision models such as MobileNet v2 at almost 400 FPS, in a power-efficient manner. In simple terms (that I can almost understand) it takes Machine Learning (ML) tasks and runs them much faster than the HA Blue can do on its own. This means my little HA Blue can do things like detect people and vehicles in a CCTV video feed quickly an...

 I'll post a longer update once I've tidied up a few things and have time to write a more complete evaluation, but today I migrated from a Proxmox VM running hassio over to  my shiny new Odroid HA Blue! The process was painless, create a full snapshot, fire up the Blue and upload/restore. Shut down the old instance, change the IP on the new to match (to avoid having to change MQTT device configs) and reboot the Blue to pick up the correct IP address. I also had to move the ConBee II dongle from the Proxmox server to the Blue, and plugged in the Google Coral Accelerator that arrived today (keep an eye out for an unboxing video). Once I'd confirmed the core functionality was working fine I installed Frigate and created a simple 2 camera config to test that it was working (and using the Coral device, which it was). Result is that I now have the start of an improved CCTV solution that detects people/vehicles and doesn't just record any old motion event.

I've been using MotionEye running in a VM for about a year now and, apart from the occasional moments when it stops showing the camera outputs (I've not had time to dig into understanding why, but it's capturing video/motion events etc in the background) it mostly does what I expected it to do. One of the things I want to experiment with is using ML (Machine Learning) to make sense of the motion events. To this end I've started looking at Frigate  which gave me an excuse to order a Google Coral Accelerator (I've just ordered one from OKdo  if you're looking for a UK supplier). At some point over the next couple of weeks I plan to get this all hooked up and test it out.

Yay, my HA Blue low-power computer arrived this morning. Here's an unboxing video to whet your appetite until I get time to try it out.