Person detection ignores detection zone,V2 detects people outside of detection zone

I figured that is why you left it out, and I am not sure that is the issue, it is just my only guess because I am not seeing any shadows or trees or anything else that could set it off.

1 Like

the AI simply does not work, the exclusion zone does not work at all for AI ,it only works for motion and not for A I detection
see the image I posted I get person detection notifications for people walking by on the sidewalk on BOTH sides of the street the entire sidewalk on my side of the street is greyed out only the porch is selected and the sidewalk on the other side of the street is no way near any un-greyed out squares there is obviously a bug somewhere , either in the FW for the camera or on the server end.

1 Like

I have had that corner blocked for a few days, and I think I’ve figured it out.

I think ANY motion anywhere in my motion zone triggers the recording (as it should) but then the camera reports EVERYTHING in its entire field of view, not just items in the motion zone. So, it is triggered by whatever in the motion zone, then reports the cars parked on the street.

I kinda wish it would just ignore those and only report in the motion area, and never report anything ever in the grayed out area.

yeah but I am specifically speaking about person detection which gets detected totally out of the detection zone ,not motion

And I have noticed that when I get the person detection out side of the detection zone people walking by on the sidewalk are outlined head to toe with the green rectangle in the person detection video clip, not sure if it is just supposed to do this no matter mater what or if means anything , maybe when it does work normal it sill outlines them even if they are outside of the detection zone ,
also I am pretty sure I never had these issues with the old V2 FW’s , the ones that had the draggable outline instead of all the little squares , it seems like this issues came along when they switched the way the detection zone is selected , not sure if it popped up on the first version that used this new selection method or shortly there after because i had been using the RTSP FW for a long time with no issues until they dropped support for it and broke voice assistant integration
that is when I switched back to the latest version of the main fork and that is when I started having these issues.

The detecction zone doesn’t work for me either. Something just isn’t right with it. In playing around with it for hours and hours over weeks and months since I’ve had the system, Ive just given up on trying to set a detection zone . In one test I litererally had ONE pixel highlighted as my detection zone and it still notified EVERYTHING that walked,crawled, blew or drove into the cameras frame of view. So, it just doesn’t work. . Ive given up and just use the system to record every waking moment figuring that if something happens I’ll have to go through all the footage and find the event. But that’s pretty much what I had to do before since it was triggering to everything anyway. Even with sensitivity settings turned down to just about nothing, it still alerts to EVERYTHING on screen, ignores the selected frames And this is on all 5 of my cameras, inlcuding the doorbell one. It’s frustrating, but it is what it is. You get what you pay for. These are like $40 now or something :man_shrugging:. If you want something that works, I guess you gotta pay for it. Oh well. When they fail, I’ll go with something else on my rental properties.

1 Like

Yeah well it used to work great , i would still probably work great if you flashed an older FW to the cams but then you would not get the voice assistant integration working because it is no longer supported with older FW
I am getting really sick and tiered of these big tech companies constantly changing their API’s
constantly , they never add anything useful or make anything better , they make it S itier and S itier and just break things and remove functionality , Google is the absolute worst when it comes to this S , but amazon is quickly catching up , every time they “update” their API some of me devices stop working and I have to go out and buy all new S
some times there is not even any option for a workaround or replacement either it’s just permanently broken and there is nothing you can do about it
Google Amazon and Samsung have all done this , nearly every smarthome product I have bought as ended u- in the trash after not even 2 years because of this crap I am really getting sick of it , Hue bridge, smartthings hub , broadllink RM pro , lots of devices that worked with ST but no longer do , all rubbish now.
There really should be a law that says they cant sell this garbage and then later remove functionality or render the device useless through “update” or server side changes" they should not be allowed to push "updates unless they are confirmed to not break existing functionality because this ALWAYS seems to be the case
buy a device , then they slowing break it and remove features with each successive “update” until the thing is not longer capable to doing what you bought it for in the 1st place just outright doesn’t work at all anymore.
There is no backward compatibility with these API "updates which really pisses me off to no end
if Amazon changes their Alexa API they shold make sure t does not break compatibility with anything first , like the old Wyze cam FW’s they should all still work even if the API changes but they don’t so we are forced to update the Wyze FW to newer broken one with less functionality , no RTSP and no working detection zone.

New to the Wyze hardware and I’ve noticed as well that the AI doesn’t respect the detection zones.y garage cam is always sending notifications for vehicles despite having the entire road blocked off from detection. It’s annoying, not the end of the world, but I’m sure the Wyze developers are aware of the issue. It’s just a matter of ensuring the software respects the detection zone when using the AI. If I had to guess they are likely aware of the issue, have it in a future sprint to address, but likely not top priority as they’re likely trying to roll out other features.

But, if a developer is reading these comments, then I’d suggest that anything with “AI” in the feature name is going to be scrutinized heavily by users given that it’s “AI”. So if you want good reviews of your software you should ensure that any defects/bugs are addressed in upcoming sprints. Add other features after your defects/bugs are addressed. I would submit a feedback specific for this issue in the app, but the AI option says “coming soon” for feedback issues.

Overall I’m impressed with the Wyze products so far that I have and I’m looking forward to improvements in the AI.

I have the Wyze Doorbell Pro with Motion Detection zones (and Cam Plus) and it works like this:

You can calibrate the [Motion] Detection zone, this is set locally on the camera to ignore parts of the frame for when it starts/stops detecting motion (sending video to the Wyze serves for analysis).

The Wyze servers then perform AI detection on the entire frame (full video) sent, to detect and notify you for any Person/Car/etc found.

Even small lighting changes in the motion detection zone may trigger the motion detection. This is a problem at night because car headlights beam is strong and wide enough to alter ground lighting and trigger motion detect.

If a person or car happens to be in view when motion is triggered CamPlus AI will detect and notify.

If you have Cam Plus and your camera has a road in view i recommend turn off Vehicle AI - and set only notify for AI events. That way, cars will be quietly recorded as Movement. But People AI will notify you on your phone.

I recall an article that the AI detection used to be constrained to the motion detection zone, but that is no longer the case - probably because for most recordings a person would be much taller (outside) the motion zone.

The AI used to scan the whole frame, but now it respects the detection zone. However, if any of the person overlaps into the detection zone, it will detect the whole person.

In m
y case there is absolutely no overlap at all it straight up detects people who are 100% outside of the motion detection zone no matter what (V2)
See the photo I posted ,it is detecting people walking by on both sides of the street

@IEatBeans on my Video doorbell pro the AI detects Vehicles (and people?) outside the detection zone with no overlap, after a motion trigger inside zone .

Do you have a link to the announcement for this and which devices currently?

I don’t have the link, but in an AMA with the AI team they said this was implemented, but didn’t say which devices. I’m guessing all, since it’s probably a cloud change.

Can you send a screenshot of your detection zone and share an event video that was incorrectly tagged

EDIT

You said video doorbell pro. I’m not sure exactly what one has this issue, but I have read about detection zone issues on one of the doorbells. It may have been fixed in a recent beta? @SlabSlayer would you know any more?

1 Like

It is the VDBv1, not the pro, that had detection zone issues in the FW. That issue is currently being tested in the new Beta FW.

Ok, so that was the one that the detection zone has to be rotated or something.

Which was the one that has the stuck on 3/3 issue?

Sorry don’t have any Wyze doorbells so am not keeping up with them :slight_smile:

Both issues same VDBv1. Current Beta FW in testing now.

1 Like

So, here is an overlay of the selected detection zone and the triggered image. It does appear that this software includes the bounding box in its detection algorithm even though the amount of movement is insignificant. Just for reference, I have my sensitivity set on my v2 to 77. I have no clue why the bounding box is so high but I would expect that the small portion that is included in that should not trigger an event.
Screenshot_20221104-231054_Wyze

The green box has nothing to do with recording or events, it’s generated by the cam and is just there to be seen.

If there’s any overlap at all of motion, it will trigger it. I would leave as much room between the movement and the detection zone edge as you can

Can anyone confirm or deny that AI is now detection zone bound?
I have had problems with all (5+) of my v3 cameras not registering AI detection for months now. The hit rate was maybe 1 in 5, usually 1 in 10. Some cameras would never register a person. Ever. In all cases motion was detected just fine.
I recently opened up the detection zones on a couple of cameras to include more of the screen that would have a person in it (think head to toe) and presto! Now ALL of the cameras are recording AI events correctly and at a high success rate.
Not sure if I like this “feature” or not. I would really like the option of AI being bound by the detection zone or the whole frame. I can see use cases for both.

Yes, the AI respects the detection zone, but if there’s any overlap it will scan the whole person.

A setting for this would be nice

1 Like

Every time I do a controlled test on my V3 after FW updates the results come back very positive. The AI will not tag me (PD) unless I personally produce motion somewhere in the included DZ (hand, foot, full body, etc).

If motion is created in the included DZ by some other object and I am fully within the excluded DZ, it will not tag me.

What I am uncertain about is on what cams this DZ AI restriction is operating.

Case in point my Video Doorbell V1 w\ FW 4.25.1.314

Today was a windy day so I did expect a few additional motion events. What I didn’t expect was 12 false positive Vehicle AI tags.

Here is my DZ (sorry, don’t have a day screenshot)

However, every one of my false Vehicle AI Events was caused by this (or a car moving on the road) clearly fully within the Excluded DZ.

This leads me to suspect that the limitation of AI tagging to within the DZ is not working on the doorbell.

1 Like