Here is a photo of a waving flag. I’m including in the hopes you’ll use it to train your AI what it looks like. It does not look like a person, a vehicle, a package, or a pet. Godspeed.
If you are getting incorrect AI Tagging on you videos, it is best to use the Thumbs Down icon and submit the correct Tagging to Wyze. Below is a screenshot showing how Person is Tagged (along with the Yellow circle around the Thumbs Down icon.
Here is the video clip that AI thinks something in view is a Person.
I did use the Thumbs Down icon on the above video to submit the correct Tagging.
Random thoughts:
Sometimes the flag is just the event trigger but the detection is something else in view such as a mailbox or different shadow patterns.
However, a flag pole often tricks the AI too since many of them have a long body with a round ball at the top that it thinks sometimes resembles a human outline with a head at the top. Even a droopy flag can sometimes look like people’s baggy clothing in the wind in some cases.
They definitely need a lot more examples to train the AI to ignore flags. It’s a common criticism. Everyone with the issue should keep submitting examples from the various camera models and various angles and conditions. Getting a lot more examples for training on false positives would definitely help. But they need to be from the cameras themselves. Trying to train on other perfect pictures of flags not from the Wyze cameras might actually make it worse when the view is actually different.
I would say maybe it could be trained to generally ignore the red white and blue pattern, but there are people who wear such patterns on holidays, etc and we wouldn’t want to turn that into camera detection camouflage either.
On some camera models it can help to set up a detection zone to basically block out the flag so it ignores the flag and flag pole from detections. That seems to work well for some camera models and not as well for others, but it can be worth trying to reduce false positives.
That’s what I’ve tended to do with different video doorbells, and that’s been reasonably effective. 