For some time now, I’ve been trying to set up a local AI-based video surveillance system using Home Assistant running on a Raspberry Pi, together with Frigate as an NVR solution. The idea was simple: offload object detection to a Google Coral TPU to reduce CPU load and improve performance.
On paper, everything looks straightforward — install Frigate, configure it in Home Assistant, add the Coral hardware accelerator, and everything should just work. However, in practice, I ran into issues almost immediately.
What I did
My Home Assistant instance is running on a Raspberry Pi, and everything else works fairly stable. I installed Frigate and started with the basic configuration. The goal was to enable the Google Coral as the AI accelerator for object detection.
But the problem started right at device detection.
The issue
No matter how I configure it, Frigate cannot “see” the Google Coral. The accelerator simply does not appear in the available options, or it fails to initialize.
I tried:
- checking the USB device connection
- restarting Home Assistant
- reviewing the Frigate YAML configuration
- inspecting logs for errors
But the result is always the same — the Coral is not detected as an available accelerator.
Current state
Home Assistant runs normally on the Raspberry Pi, and Frigate starts without issues, but everything falls back to CPU processing because hardware acceleration cannot be enabled.
In practice, this means:
- higher CPU load on the Raspberry Pi
- slower video stream processing
- no benefit from the Coral TPU as intended
Conclusion
For now, I’m sticking with CPU-based detection until I figure out the root cause of the issue. It could be related to drivers, USB detection problems with the Coral device, or the Frigate configuration within the Home Assistant setup on Raspberry Pi.
If anyone has a similar setup and knows what might be causing this problem — any help would be appreciated.