Motion Flood Lights and False Person Detection

OG Wyze user here, first post. I imagine this has already been covered, but I couldn’t find my specific issue. Most of my cameras are V1 or V2, but I got a V3 to monitor my detached backyard office, arguably one of my most important cameras because of the seclusion and how much money (PC, etc.) is in there. I have the camera under some old motion sensor flood lights arranged similarly to the flood light package that Wyze sells. I got the recording zone dialed in pretty well so it’s not picking up passersby outside the yard, and I’m sure upgrading to Plus would stop my dog from setting it off, but many times when the flood light turns on it sets off Person Detection. Any tips?

@topherludlow, welcome to the community!

I have 2 CamV3 mounted over the top of 2 old school motion sensor floodlights. I went thru considerable tweaking sessions to get it right and it is still not perfect… Also be aware that the current rate of false AI positives for the V3 is very high right now, regardless of the placement. The pixelation detection algorithms need some serious tuning. You will find copius forum posts on that. Currently, cars throwing headlight beams will set off AI to a false positive. It is because Wyze AI looks for the volume of individual pixel changes in light over a short period to identify objects. If your floods are on and the cam is in night vision mode, trees and bushes moving will initiate an AI positive.

Because my floodlights are PIR sensor activated, when my cams went into Night Vision mode and turned on the active IR emitters for the night vision, it flooded the detection field with IR and set off the floodlights. Because that caused a massive light pixelation change in the cam when the lights came on, the CamPlus AI always tagged it as something. At first, they fought each other all night long because the IR emitters would go off after the lights came on, the timer on the lights would shut the lights off, the cam would go back into night vision and the cycle would repeat.

Over time, I was able to get the lights to play well and get along with the cams by setting my cams to NEAR on the IR settings so they didn’t emit so much IR and dialing back the sensitivity of the floodlights PIR sensor. I also lowered the detection sensitivity on each cam to reduce false positive AI Events.

It took time and a lot of testing and monitoring, but it eventually calmed down to an acceptable level.

Hope this helps.

1 Like

Screenshot_20220306-103550
The answer seems very obvious to me. The algorithm thinks this is a person in the frame. (It doesn’t matter if it’s outside the detection zone.) Your only options are to keep filing reports against it, or to move it or the camera view to exclude it…

2 Likes

Thanks!
The description of that sensor loop is pretty hilarious, though I’m sure it was frustrating. Fortunately mine isn’t nearly that bad. That being said though, I’m disappointed that it’s having issues considering it has probably some of the easier conditions of the bunch. A couple things you mentioned… You said v3s are having issues. Do you mean they’re having more issues than the older cameras? I would’ve assumed their algorithms would work better and better with camera improvements, but I guess you’re saying the increased camera quality actually complicates things? Would swapping that one out with an older one help? Also, do you know if the flood light combo they sell works any better? I can’t see how it would, but maybe they’ve optimized it somehow :man_shrugging:. To be clear, I’m not having the same kind of issue you described where the light sensors and camera IR are interfering with each other, they’re pointed in different directions.
Thanks for the help!

When you say detection zone do you mean the green box indicating what’s setting it off or do you mean the “detection zone” where you set what areas do or do not set off the camera? In either case, I suppose I could just throw a trash bag over it and see if that helps.

Yes. Every time Wyze does a Firmware Update on a cam type, in this case the V3, something inevitably goes haywire. I don’t really know why, I am not a programmer or developer.

One of the recent updates included some change to the AI that degraded both the AI accuracy and the positive notification rates. I am told by Wyze that they are aware of the issue, particularly on the Doorbell, and are working toward a fix and update.

No clue. I have V3, Doorbell 1, and PanCam V2 (indoors when away).

Don’t have one. Don’t know. Check the forums here and find a @Mavens or @Mods that has one. Tons of posts if you search for it.

I’m not sure what @customer was referring to, but my doorbell is definitely ignoring the detection grid zone set manually in the detection settings in the app for space to be ignored. It is giving AI positive notifications and events clearly in the blackout zone. The green box you see moving in the video is a movement tracking box and only serves as a visual reference for you to see what it is that caused the AI event being recorded. This can be turned off easily. But, it is useful when it is tracking movement for an AI event that is clearly outside the Detection zone.

I mean the latter. Wyze person detection flags when there is a person anywhere in the frame.

Yes, covering the object would be a good test at least.

I believe the following was meant: When motion is detected in an area you said to monitor, the entire stream is sent for interrogation. So something occurred which decided this is an event you would like to know of, the stream is sent in for interrogation and the entire frame is reviewed with the algorithm and not the selected area. The Image then looks like a person and you are notified.

Another way of saying this: The Selected Detection Zone is used to determine motion. The entire Frame is used to determine if any of your AI Selections exist for notification.

Based on what was indicated, the light would indeed trigger an event of sorts as there is a change in the image. The Entire stream is sent to the cloud for interrogation and it sees the White statue, bird bath, or whatever it is, as a person and notifies you.

2 Likes

Do you have any insight as to why I am getting events for motion that is happening in the blackout area of my detection zone? I understand that the AI looks at the entire frame, but if the motion being tracked is in the blackout grid, it should have never initiated an event in the first place.

Did you upload the video of the event? If you have it, can you upload it?

My belief, based on the image above, is that your light went on. The change in lighting actually triggered the event. which then triggered the notification after interrogation.

If any changes are detected in one of the Detection Zone Squares you will be notified even if it is one pixel within that square. In the image, the light in front of your office area is on - well lit by the way :), any changes of light will trigger the event.

You can test to see if the white statue is causing the Person detection by covering it for a night and then trigger the camera by turning the light on yourself (if possible).

No, @SlabSlayer is not the OP who posted the statue/something screenshot.

1 Like