Just set this up to try and recognize people from my unifi protect cameras, I want to replace the large number of person detected notifications I get from their system to something a bit smarter that won't spam me every time I step outside, I'd like to ignore known people and alert only on unknown faces.
The G4 Pros already do person detection, so I created an automation to run image_processing.scan on cameras reporting a person in frame every second until their detected object entity goes back to none.
I then went and walked by one and saw some image_processing.detect_face events flow in : great.
Now someone else just walked by a camera, and the automation did trigger (a lot) of scans, but no events got generated.
Looking at the recorded footage I suspect the issue is simply the angle, the person's face was almost always facing away from the camera which unfortunately is going to be true a lot of the time with this.
Is that correct, running a scan on a picture with no visible faces does nothing at all ? If so I'm thinking a separate event could be triggered when the scan completes saying no faces were found, that might be handy.
If there is indeed nothing returned when the face isn't visible then my plan B would be something like this :
One automation with a trigger on the camera reporting a person in frame, running a deepstack scan per second
Another automation with a trigger on the camera reporting a person in frame, with an action to wait on the image_processing.detect_face event with a timeout.
If it gets a known face event, do nothing
If it gets an unknown face event, notify our phones that an unknown person was seen + picture
If it times out after X seconds or minutes with no events, notify our phones that a person was seen but no face could be found + picture.
A 'no face found' events could be nice here to avoid relying on a timeout, but that should be fine.
From the perspective of someone who's been using this integration for more than 5 minutes, does the above make sense or am I missing something ?
Hi,
Just set this up to try and recognize people from my unifi protect cameras, I want to replace the large number of
person detectednotifications I get from their system to something a bit smarter that won't spam me every time I step outside, I'd like to ignore known people and alert only on unknown faces.The G4 Pros already do person detection, so I created an automation to run
image_processing.scanon cameras reporting a person in frame every second until their detected object entity goes back to none. I then went and walked by one and saw someimage_processing.detect_faceevents flow in : great.Now someone else just walked by a camera, and the automation did trigger (a lot) of scans, but no events got generated. Looking at the recorded footage I suspect the issue is simply the angle, the person's face was almost always facing away from the camera which unfortunately is going to be true a lot of the time with this.
Is that correct, running a scan on a picture with no visible faces does nothing at all ? If so I'm thinking a separate event could be triggered when the scan completes saying no faces were found, that might be handy.
If there is indeed nothing returned when the face isn't visible then my plan B would be something like this :
image_processing.detect_faceevent with a timeout.A 'no face found' events could be nice here to avoid relying on a timeout, but that should be fine. From the perspective of someone who's been using this integration for more than 5 minutes, does the above make sense or am I missing something ?
Thanks !