Now that person detection is implemented. Facial recognition could be next? In that, the camera could recognize a face and expose for only that area. I’ve caught thieves on my wyze cam but their faces were still unrecognizable.
I believe that this is a hardware limitation due to the camera quality. The only way to get facial recognition would most likely be a better image sensor. Hopefully the outdoor cam will have a much better image sensor to make facial recognition actually work.
I think that is a great idea! Another possibility would be adding an option of “Person Detection” under Event Recordings, to insure that the camera ONLY records Events that include “Persons Detected”, as opposed to any type of motion which now includes leaves and/or shadows moving, lights turning on and off, etc. In my case, I usually get 1 or 2 “Person” events, and I end up having to delete between 20-30 “Motion” events from the different cameras.
I think this is a case that the normal motion detection is implemented in camera, but then the clip is sent through to Xnor.ai to analyze for people and then it is flagged. So you basically always have to record with normal motion first before Person detection can work.
Then again, I supposed they could just wipe the non-person clips afterwards.
Welcome to the community, @ebendl. The xnor.ai person detection processing is done directly on the camera hardware.
Really? OK that’s pretty great!
I probably should have included this link the first time.
WYZE FACE Recognition:
Your ticket (338598) has been updated:
Where are you with the above feature?
We are a non- profit thrift store and have folks coming in all the time and really would like to be alarmed if certain people come into the store. Possible know shoplifters. Thanks to WYZE videos.
We have a perfect face picture ( We can almost find dental records from the picture :- ) ) of selected persons as reference, so it’s just a matter of matching incoming with those reference face pictures.
I also see mention of picture qualities in this feed. Its all about lighting and optics, even IR cameras have a problem recognizing car tags due to reflections. from plates.
This, I believe, is going to be hard to do (specific facial recognition). As more cities and states make facial recognition illegal I doubt companies that sell everywhere will implement something like this. Just my guess though, I do not work for them or have any special info.
@Jason21271 Thanks for feedback.
It is just automating the current manual process of a person viewing 24 hours of video manually, which to my knowledge is not illegal. Especially if you have clearly visible notices all over, that the premises are under 24 hour video surveillance…
Alternative is WYZE adding 2x, 4x, 8x, etc. fast forward and fast reverse playback feature, so we can do the manual process faster.
I am definitely far from a lawyer and my understanding is manually doing it is fine but if you have a processor do it its illegal, but again not a lawyer and have not read the actual laws since they have not affected where I am yet.
Judging by the heat they’re getting on their own Facebook page… They’re going to regret that move. #sorrynotsorry
(post withdrawn by author, will be automatically deleted in 24 hours unless flagged)
There may be some Smart Life compatible cameras that can do that for you. Check out this link.
Have pets in the house… Need person detection AI ASAP.
They are definitely working hard on this and making good progress. However, at this time, there is not a firm schedule for release.
Notice of recent removal of a third-party person detection capability indicated Wyze pursuing alternate design/AI. Thinking about this, I want to offer my thoughts. Reviewing many of the ‘person’ triggered snippet videos that had no person, it was is noticed to be somewhat complex motion that obviously confused the AI. (E.g. trees flowing by wind, shadow of the tree flowing by wind, a flying bird shadow, occasional vehicles accelerating, night vision transitioning on as lighting changes or from lights turning on). I appreciated this feature when it worked…highly noticeable now that it’s gone.
Wouldn’t it be valuable, since there are infinite set ups, unique camera angles, daily repeated motion such as same tree blowing or shadows from tree, spiderwebs reflecting as wind blows triggering night vision lighting etc… to allow a user to respond affecting types of desirable recordings of motion and earmark (machine learn) those a user defines as not useful. When a person is noted in a motion snippet, user marks as person motion. Other motions marked by drop down menu selected and able to remove from triggers…although a user can review machine learned /removed set of videos to confirm or deny that the machine has learned properly what types of motion he does or didn’t want to exclude…thus continuously improved machine learning.
Now that I’m having to constantly review motion, this has become very clear how valuable only certain snippets are. I’m very interested to discuss further if explanation isn’t clear above. Thank you,
With my primary cam on it’s side for monitoring vertically for coverage of the front door entry from head level to floor, it rarely detected a person so hopefully wyze takes this into consideration since I have read of many others in the rotate 90 deg wishlist thread using it in a similar situation.
The problem Wyze has is the anemic CPU and the very low amount of RAM they have to play with.
Having lots of different models would be required, and they just don’t have the ability to store all of them on the cam.
If they switched to a cloud processing, then, yeah, this and many more things can be possible, but, obviously, it would cost them much more.