So there’s no point in paying for it.
And Wyze claims it’s learning and why we submit the corrected label to them. It’s been years now and I have seen no improvements, at least with my recorded events.
Concerning AI issues . Had to disable package detection on a front view camera because it starting going haywire sending package detection on front steps railings with solar post lights and a tree branch waving. Then I got person detection notifications on a moving tree branch shadow on a gutter downspout. Reported the videos but no category for shadows or structures that do not fit.
So tired of getting notifications for the vehicle parked in my driveway but I can’t turn that off because I won’t know about the person walking up my driveway. Just had an Amazon driver walk right up to my front porch and put a package down and I got a vehicle notification but nothing concerning the Amazon driver!!! I’ve submitted so many corrected videos but it never changes 
I’m cancelling my subscriptions, total waste of money.
@samtron22001, you customize AI detection in event recording and notifications. I have a vehicle parked in a detection field of a camera all the time. You simply uncheck the vehicle boxes and no more notifications. Had to do the same for package detection on a couple cameras. Works flawlessly for me and does not interfere with person detection. You do have Cam Plus now Cam Unlimited I assume?
Looking back at previous conversations, I have not had the kind of issues you have experienced. Never a dog or animal as a person, always a pet. I have had shadows, tree branches waving and an occasional inanimate object falsely reported as person. I would like to see Wyze expand the list of what you can report in a submitted video or give you the ability to identify ‘Other’ in a submission.
But when you uncheck Vehicle detection then you also don’t get moving vehicle detection and maybe someone wants that, they just don’t want parked vehicle detection. I want people, vehicles and pet detection on most of my cameras but only if they are moving. When I set it the event recording to smart detection events, it will often miss those things.
But what really bugs me is that a while back Wyze reduced the recording sensitivity with an update and now my cameras miss even more things that a few years back they used to be able to record.
Right 
I get notifications of a parked vehicle in my driveway when the camera is triggered by a person but there is no notification of the person. Ridiculous. The moving person triggered the notification, not the parked car. If I turn off vehicle notification, I miss the person coming up my driveway. It does work some times, but the system is very sporadic.
I have an update. I reset the cameras services in the application 2 days ago and I have vehicle notification turned off, it seems to have made a difference. I am getting notifications for people when a person triggers the camera. I will turn vehicle notification back on and see how it goes.
I’m not really seeing a difference. A tree limb blowing on the left side of view can trigger a recorded event and it will label it as a vehicle because on the right side of the view is a parked car.
I have cam unlimited and a V3 pointed at a large front driveway. I was careful to not point it towards any trees or bushes. However, in the morning, shadows from palm trees on the east side trigger events. It tags the vehicles parked on the west side. Turning off vehicle detection is not a solution because I want to know when a vehicle pulls into my driveway.
I see a lot of complaints about parked cars causing false events.
It should reasonably be expected that if my camera is pointed at the driveway, my parked car should not trigger an alert when a tree waves, or in my case causes a shadow in the driveway.
It sounds like the problem is that the camera, when uploading the video to AI, doesn’t include the area that triggered the alert so it can determine if it is valid. This would be an application design flaw.
If you have motion tracking on the event will have the area marked and it will be shown on the playback. Cloud and sdcard will both reflect as such, at least in my experience. You may need to increase motion intensity. How do you tell AI to ignore something it is requested to show. Distinguish still from moving? I would like to see Wyze engineering tackle that algorithm sequence. Good luck with that.
I’ve been saying for years the Wyze AI should only label what moved, because I don’t care about parked cars! I want to know what triggered the event. The tree waving trigger is a huge sore spot, I try to block them out with the detection grid but the grid boxes aren’t small enough to fine tune it.
I usually have a couple hundred events to review daily and when the AI sees parked cars when people or pets triggered the events but it didn’t label them, then I end up having to review every single recording because I can’t trust it.
Well again, it should only label what actually caused the motion event. Why would anyone want to know about a parked vehicle?
Nobody wants to know about a parked vehicle and I may have replied to the wrong comment; what I was meaning to reply to was an earlier discussion surrounding detection zones, and how that can impact object/AI detection.
This screenshot is obviously not Wyze but it doesn’t matter, ignore that for now and let’s just pay attention to the principle pieces involved.
Let’s say I only want to see motion in the area (sloppily) highlighted in yellow. I only want to see things on the driveway and yard and sidewalk, but nothing in the street. So I set a detection zone as such. Problem is, object detection models work with polygons, rough, jagged polygons with right angles. Not the smooth outline of the car, but the whole box around the car.
This is why motion zones impact AI object detection
As you can clearly see the car is not on the sidewalk; however, for many detection models it IS. Since the detection happens in a box, and that box happens to be on the sidewalk, for all intents and purposes, that vehicle is on the sidewalk, it’s actually in the yard as well.
To combat this, when running inference in the detection model, we can use logic like:
-
Contain: the entire object must be contained entirely in the zone
-
Intersect: the object can just “touch” the zone
-
Exclude: a great example of this use case would be setting an exclude zone on your neighbor’s property that’s visible in your camera view. This way, you’re not constantly getting alerts about their comings and goings.
-
Include: an example being your driveway. You definitely want to know when something’s there.
Combining parameters is where things get really interesting. Here’s a cool trick: use Exclude + Contain in one zone. This creates a dead zone where nothing fully inside it will trigger an alert, but stuff partially in might still set off other zones. It’s perfect for things like busy sidewalks or that annoying tree branch that keeps setting off your camera.
I’m not at all up to date with what Wyze let’s you configure, I was simply trying to help you out with your question below. Not in a wyze-specific sense but a general sense, of what is happening. It is much less frustrating when you understand how thing work (and when you understand how things work, you will understand that AI is not magic, and you should be patient : ) machine learning has got a long way to go, a really long long way to go, with AI Vision Models
I just thought of an even better example of combining logic and zones: Package Detection. Here’s the kicker - the best way to detect a package doesn’t involve image recognition at all.
Stay with me here. Imagine this scenario:
- A doorbell camera detects a person on the front porch with high confidence.
- That person then leaves the frame.
- After they’re gone, there’s a new, stationary object that wasn’t there before.
Boom! That’s a package 99.99% of the time. No fancy image recognition needed.
Think about it - packages don’t magically appear without human involvement. And people don’t usually show up on your porch just to leave random objects behind. It’s almost always a package.
(And before anyone chimes in, let’s set aside drone deliveries for now.)
I’m not up to date on Wyze’s AI capabilities, but I’d assume they’ve incorporated this kind of logical approach. Scrypted has been doing this for a long time.
Im getting alerts for a person but no person, an insect. Rain, shadows from cloud which happens a lot, insects trees swaying, you name it, it triggers a motoon alert and the odd time a person alert. Ive fone all the troubleshooting and reset but no joy. Very frustrating. Was a tester for their new ai search which didnt work for me either. Imho the term AI and wyze do not ho together.
Since the AI isn’t smart enough, or I should say Wyze isn’t smart enough, to give us AI detection only within the grid we set then I would suggest they give us even smaller grid boxes to fine tune our detection zones. I think that would solve a lot of our false detections. For instance today it was windy again and tree branches with its shadows kept initiating a smart even of a vehicle that was parked across the street. I had to try to block out that car which was hard because the grid boxes are too large and it then blocked areas I wanted covered. This was difficult because the camera doesn’t have a straight on view of the street. I think the grid boxes need to be at least a quarter smaller than they currently are.
That won’t change anything. This is not a wyze thing. This is about how object detection models work, like, essentially all of them. if the boxes were literally individual pixels, this car would show up in the lawn:
Now, that being said, there are other ways, as I stated before.
If you add your camera to a platform like Scrypted, that gives YOU THE CHOICE over how zones interact with objects/motion/detection, you can say:
- I want anything on the sidewalk to be a
containzone, NOT anintersectzone, which is likely what you have now (we don’t know, of course, wyze doesn’t use open source models, scrypted does though) - When it’s set as a
containzone, the whole detection box must contained within that zone. - Yes, if you’re wondering, that does mean a car could drive up into your lawn just slightly, but still not be fully “contained”, into that yellow zone
- But how often does that happen lol?
this is where better camera motion detection comes in, if you want to take it to the next level: https://youtu.be/6P9v6l892OU
you need a cam with great motion filtering. That link (it was too big to upload) is an insane storm, and what you’re seeing on the right, is an-almost-blank timeline. Because it only shows events that is actually important; only relevant motion is sent from the camera (that is an Axis camera and they are expensive for a reason, but can be found quite cheaply on ebay, when a school district, for example, replaces all their cameras; mine came from the DOT).
There are a lot of ways to solve or at least mitigate your problem, but you have to understand, it’s not a wyze problem, from an AI perspective.
It IS a wyze problem from a “let me configure the properties of my zones the way I want” perspective. But again, this can be solved via 3rd party solutions, both with software and your wyze cam, or most ideally, at some point, a new camera, plus scrypted, or frigate, or another open source solution.
