Pet detection (events) works mostly. Problem is it will detect a 3 pound cat (as an event with notification) but not a 30 pound coyote (no detection at all). Have 2 cameras in back of house watching back yard and ally (dual pan cam and a V4) both will detect cats (mostly) and notify me. Had a large coyote jump fence from ally, then run to patio on side of house. It ran past both cameras without triggering detection (set to high). Watched out window, then checked SD card and coyote was there, no event or notification. Explain, please fix.
Most of the time my V3 camera won’t even detect a large dog but it detects the person walking it. And it always detects parked vehicles, yet often misses moving vehicles. Also with I’ll get 3 notifications with every event. Often I have to turn off notifications because with 10 cameras they bomb my phone to death. Wyze has been working on the software for many years now and haven’t really fixed the things we needed fixing. They finally added Birds even though it rarely tags them as such. Insects and Tree Waving have never worked on my cameras, yet I get a lot of motion from both events. Not sure why they added Face, Package Delivered or Package Picked Up, when Person is all we needed.
And for a security camera all that is important is what moved, not static objects in the background that got tagged because the wind blew. They’ve been talking about only tagging what is in the motion box but apparently that’s outside of their skill set. They also should increase the pixel sensitivity because they reduced it years back causing my cameras not to record events it used to be able to easily tag as motion, now they don’t see it even though it’s clearly identifiable in the a live view.
I have Cam Plus Lite and only get Person Detection. I have found my v3 cameras with Detection Sensitivity set at 90 aimed through double pane windows to be mostly accurate at recording Motion and Tagging a Person. I do get occasional False Person Detections mainly at night. This Cloud Event clip recorded a person across my street and Tagged it as Person.
Did it also tag the Dog as a “Pet” at the same time?
SJ
I have Cam Plus Lite, so I only get Person Detection.
Wyze Ai detection is a colossal failure. Leaves detected as a person, reflection of the wyze doorbell light on brick as a person, light filtering into house as a person, ridiculous
FYI, these things are usually just the event “trigger” and not actually the thing on screen that was detected as a person. There are a lot of reasons I know this, not least of which is because I asked about it in an AMA and one of the AI Product Managers responded and publicly offered to push a special firmware to myself and a few other people where it would highlight in blue the actual AI detection, and not just the motion. So I got to witness with total accuracy that things like leaves and light, etc weren’t usually the things being detected as AI objects and that they were usually just the motion event trigger.
Additionally, my experience is matched by the other brands’ cameras I have that often misidentify things with false positives as well. I had eufy cameras that would frequently misidentify chairs as person detections. Then if I would change the position of or add things onto the chair (like a blanket or towel, etc) the false identifications would go away. I have done similar things with other objects and brands and helping other people find what is actually being falsely identified.
If you’d like some of us to give you some ideas of what are the likely false positives in your camera view, feel free to upload a sample video or screenshot we can analyze. Some of us have gotten pretty good at identifying the object being falsely identified and good solutions (sometimes slight adjustments to the object or camera detection settings can fix the issue). Also, sometimes if you submit enough videos to their training data, the issue fixes itself over time, but they only update their AI training data once a month, and it can take a few months to start noticing a difference in certain cases, but I have personally seen this go from nearly 100% false detection at night to basically 0% after I submitted enough samples.
Sometimes you can just block out a couple of grid squares so the object in question is no longer analyzed by the AI.
Anyway, totally understandable if you don’t want a bunch of internet strangers analyzing your event video, but the offer is there. Just know that there is probably something else in view of your camera getting identified as a person, and it’s usually not the things that are actually triggering the event. Also, it can vary a little depending on the camera model. Some cameras have an upgrade that tries to ignore stationary objects, while others don’t seem to.
Thanks for the reply. Unfortunately I can’t move the tree or the brick on my house. I suppose I can black out all my windows to prevent shadows. I have been submitting the feedback ever since that was an option and I have been submitting a lot but to no avail. There is nothing else in the video other than what I listed, yes detects motion than it’s as a person and yet when a real live person shows up there is a delay. 5 mins to 10 mins, sometimes longer. My son has eufy cameras installed at his place and does not experience this nonsense. Like I said I submitted feedback, a lot, and nothing has changed
I could block out a couple squares as suggested but then cameras would be useless
Thanks for sharing. I can see your confusion. You are correct that the tree leaves are absolutely the thing initiating the event because of motion. That is why the green motion box highlights the leaves like this:
This leads you to understandably assume that the “Person” detection is the thing inside that green box. But that is probably not what it is being detected as a person. I would put the probably of that down to no more than 20%, but it is definitely a possibility. vague anthropomorphized shapes like this are a possible false trigger:
BUT it is fairly rare with leaves because of all the holes instead of being more solid. Not impossible, but one of the less likely explanations. Still, you could block out parts of the tree that could have vague human shapes if you squint and turn your head. But I don’t think that’s the regular false detection. I’m pretty certain that is just the motion trigger.
The real thing being falsely detected is something else. When the branches sway, the entire view gets analyzed by the AI, including things that are more stationary. I’m pretty certain it is this plant and stool being falsely detected as a person:
It is not the thing triggering your events with motion most of the time, but it is what likely keeps getting recognized as a person.
I know it’s still ridiculous that a plant is being tagged as a person, but in my experience this is highly likely to be what’s going on. This is because of
- the beautiful circular flower resembling a head shape
- It’s on a stool/chair, and stools and chairs for some reason make LOTS of companies cameras trigger as a person (including but not limited to Wyze), especially when certain things are on them.
- There are a few reasons for this, including the way they are structured to analyze shapes and outlines rather than the context of an object and that they were all created based on similar foundational training data from open-source vision libraries like YOLO or MobileNet using massive benchmark datasets like MS COCO or ImageNet. Almost all major commercial camera brands source their AI pipelines from a handful of leading computer vision developers, they share the exact same blind spots and mathematical associations. When these core models analyze an unusual angle, shadow, or piece of furniture, the probabilistic nature of AI can cause them to default to a “person” classification. I could explain in lots of boring technical detail, but suffice it to say that the fact that you have a stool/chair with something on top of it that has rough human shape features…if I was a betting man, I’d tell you this is almost definitely your false detection issue object right here.
- There are other factors that could be listed, but having resolved several of these cases myself with multiple brands cameras and having seen the EXACT same thing myself, and helping lots of other people in the past, and researching why stools and chairs and stuff with things on them tend to trigger false positives…lets just say I’m betting my money on this being the issue.
So what is the solution?
You can set a Detection Zone that blacks out that flower/stool. You might be able to just do a couple of grid sqaures to mess up the detection enough. That MIGHT resolve it depending on the camera model in question (some models have had the AI ignore detections outside the detection zone, while some models only use the detection zone for the motion trigger area but will still detect the entire screen…so it kind of depends which way it works on your particular camera model right now).
Another option that often works is to shift the position of the stool and the flower. Maybe turn them a few degrees. Sometimes just a slight rotation can mess up the false detection outline and it will stop triggering or at least reduce the frequency of the false detection. If a slight rotation doesn’t make a difference, try doing a little more. Maybe add something else in the view to screw it up. You could possibly put a towel or table cloth over the stool which will keep it looking nice with the flower pot on it while messing up the stool visual from the camera analysis. You could shift that stool and plant more than just rotating it.
If that doesn’t fix it, then as a test try temporarily removing both the stool and the flower out of the camera view ENTIRELY. If your false detections are reduced or go away entirely, you will confirm that was the issue. If they persist at the same frequency or just a lower frequency then you will know there may [also] be another object to consider and we could talk more about those…but they’re a little less likely, so for now, my money is on the flower + Stool, so I’ll wait to describe other ideas to see if this one resolves it first.
I’d love to hear an update if you attempt any or all of the above. 
Thanks for posting an actual example event. This is why I asked for one. It is easier to give you solid examples when I can see what’s actually in view. Hope it helps.
That’s what I was thinking when I saw the video, and that’s what I’d try masking with a Detection Zone to see what that does for detection results.
I like the drawings.
I appreciateyour response and analysis. The plant and stool were added there recently. The problems as I indicated have been going on for more than a year, long before. I have instant of the doorbell light reflecting off the brick of the house showing person detected. This is constant despite the numerous feedback I have given. Beyond frustrating especiallywhen there is no support from wyze. Sorry but just crap ai
The stool and plant were placed there a week ago. The problem has been ongoing ever since I installed the camera and cam plus, 2+ years
If you really want to figure it out this is what you do…go setup a detection zone. Put blackout squares everywhere that is red:
Black out basically everything but the clear areas on your porch/sidewalk under the tree. Don’t worry, anything that overlaps into the detection zone will all be analyzed by the AI anyway, so even if a person’s face is in the red area but their torso/legs are in the clear area, their whole body will be analyzed. You’ll still get your person detections for a person.
Maybe even black out more area than I show above just to start with.
Once you try this and determine that you are no longer getting any false detections, you can unblock a few more squares at a time, then wait and make sure everything is still working well, then you can unblock some more. As soon as you start getting false detections again, go back and black out whatever you recently unblocked and know that this area is one of the problems and you’ll know to keep that blocked or modify it some way.
That is the best way to find and resolve false positives.
I acknowledge that you said it happened even before you put the stool and plant there, but I suspect that at minimum your frequency of false detections increased lately because of the stool and plant. That doesn’t mean they are the first and only thing that has ever caused you this issue, but I am reasonably certain they are making this happen a lot more often recently even if there are other culprits contributing to events in a similar way. If you start with a small detection zone and slowly open a few more squares at a time you can likely soon identify for yourself what other things are contributing to the issue if you care to resolve it.