Sebastian Pietschner - Developer and Student

My personal site that I use to develop and share new skills and projects.


TwitterGithub

Frigate NVR

Sebastian Pietschner / Oct 15th 2021
2 min read

Ever since I moved into my current house and even before we moved in I've always insisted on smart home devices. The first things installed was Google Mesh WiFi and our Ring smart doorbell. But, as I've been discovering recently a closed ecosystem isn't to everyone's benefit.

Once we stopped paying the Ring subscription service the majority of its most useful features disappeared. One of those features which we used extensively was event history. On one occasion the side gate to our house was left open and the dog escaped into the neighbourhood. Using the Ring doorbell we were able to figure out who had opened the gate and when.

Not too dissimilar to this video I believe that subscription services should be avoided when it comes to IoT devices. In past posts, I've discussed my use of home assistant and Zigbee to get around Ikea's proprietary gateway.

So, this brings us to Frigate. A project created and championed by blakeblackshear on Github it brings smart home security home with AI image recognition powered by Google Tensorflow Lite. Using a simple USB machine learning accelerator card like Google's Coral it's possible to attach any of your networked cameras to home assistant along with having features like event history.

With the Google Coral, I can process up to 100 frames per second with 1080p video and, considering detection only needs 5 frames per second to detect someone means that it is unlikely that I'll ever max out the coral's processing capabilities.

Where this project really shines through is its integrations. Integration with Home Assistant's media sources makes past footage easy to navigate and using a project called doubletake I'm able to link facial recognition into the system.

Using a simple addon for Home Assistant called Ring-MQTT I can make my doorbell integrate perfectly once again without having to rely on any third-party subscriptions or cloud products. Everything stays local.

So far the setup has performed flawlessly, sending a notification nearly instantly with video of people walking up to my door and alerts sent to my phone whenever it detects someone. This has been a great solution to the ever-growing problem of cloud dependence.