TLDR: How to configure Frigate to read videos from HA storage, detect humans in the video files, sort them to matched known people and unkown ones?
Details:
Frigate is designed to work with streaming video. It can recognize people.
Streaming 8k video to Google Coral over Wifi from several cameras is not feasible from Coral performance and from Wifi load standpoints.
HA has an atuomation which records videos with people on the attached storage. It woulbe great to feed any new video to Frigate and get from it a names of persons it can detect in the video.
How to obtain so?
If Frigate cannot do this, which solution can?
The detection and recognition of objects is supposed to run on lower resolutions (that is what it is trained for). Usually camera’s like these have a lower resolution substream for this. If you set the right masks you can also prevent frigate from wasting time on useless movements. If you use the right model that can detect faces it can also limit its efforts on face recognition itself. Plus face detection runs on quite low cpu. So before you try the less obvious (and less useful) route. Frigate should be able to handle it if used properly.
And besides: storing the data won’t make it go faster. If the detection can’t keep up with real time, analysing the stream later on will have the same problem and not keep up. Plus waiting for the recording to be saved will make the recognition come to late to do ny usefull automations on it.