Wyze AI communication - March 2021

A post was split to a new topic: To Wyze developers

I would still love some clarification on the detection zone and cam plus. I have my detection zone set to my walkway up to the house only. I constantly get notifications with vehicles detected which are parked across the street nowhere near the detection zone. Is the Ai limited to the detection zone? And if not, why not?

Also, when will the customizable detection zone be available for v3? It would really cut down on some false alerts if I could reshape the detection zone.

@Michi
Unless I am wrong, and hopefully someone will correct me if I am, the detection zone is where it looks for initial motion. When motion is detected in the detection zone the video is uploaded and scanned with the AI. The AI then looks at the entire image, not just the detection zone. That would be why you get vehicle detected for the car across the street. I know people have mentioned having it only look in the detection zone but my guess is that will not happen until the grid detection zone is finished, they are working on it for the v3 currently and recently imlemented for the Doorbell

3 Likes

You’re correct I’m sure

Thank you for that info, Jason21271.

We are looking into this issue right now, It seems truly that v3 configs might has some bugs.

3 Likes

Both of you are correct. We had a fixing version of the detection zone, in which the AI will only detect the object within the zone. But due to some tech issues, the change is reverted for now. We’re working on resolving the issues to make it more stabalized across all devices.

2 Likes

Very interesting! I could see this being very difficult to implement. For example, if I have a detection zone that covers just my sidewalk (perpendicular/horizontal to the camera pointing direction), then in some cases, only a person’s feet would technically be in the detection zone. Their knees and body and head/face would be outside the detection zone. Therefore, if a person walks on the sidewalk, it seems the AI would be told to analyze only the feet, and determine this is not a person because the body and head were not in the detection zone (which the AI was told to ignore), even though the trigger (a person’s foot) was in the detection zone.

Is this potentially part of what is causing the problem with limiting the AI to detection zone analysis only?

Example screenshots of my above description using detection zone grid:

Feet would be in the detection zone, triggering an event, but the body and head would be excluded from AI detection, thus causing the AI to not recognize a “person”:

image

AI here would only recognize a “PET” because my body/head are excluded from the detection zone:

image
image
image
(The AI would still recognize my “pet” in theory…though in my case, it would probably think my black cat is a “person” but that’s beside the point here :slight_smile: )

If the AI is told to ignore everything outside the detection zone, then anyone could walk right across my front sidewalk without me ever getting a person detection because the AI would then be told to ignore the body/head which it has been trained to look for.

I guess you could try to tell the AI to check anything connected to the triggering object (my legs) and see if there is a body/head connected to that triggering object that started in the detection zone, but when it comes to a picture, isn’t EVERYTHING connected to the triggering object in some way? I guess a person on the other side of the street would have separate legs/body/head not linked to the legs in the triggering object though, so the AI could be trained to ignore the non-connected person(s).

Anyway, very interesting. I would certainly love to hear anything you’re willing share related to the issue of AI analyzing inside/outside of detection zone limits to ensure the above examples aren’t problems. I can see you have your work cut out for you though! It sounds fairly complex to train the AI this way.

3 Likes

@WyzeShawn Could you tell me why my Wyze doorbell to this day still hasn’t recognized a face? I have facial recognition enabled and it is on my front door so it sees face very clearly. Is it not enabled for the doorbell on the cloud side or is this a bug?

Even with CamPlus I have never had my doorbells detect anything but motion. No person, face or package even though we get UPS and FedEx quite often. If I use the CamPlus filter to only see people, vehicles and packages none of the doorbells events ever appear. If I view motion triggered events they show up.

I have camplus on it since I got and it detects people and motion very good. I just don’t get why it never picks up faces when it has to best view of faces over all my other cameras. Meanwhile my garage cam mounted high always picks up my Dads face when he’s in the garage.

Something is severely wrong with my cam v3s detection. It’s mounted on my second floor looking at my backyard and goes off for every bug and raindrop that goes by it. I’ve tried setting the motion detection all the way down and it still goes off for a bug zipping by. Even worse is that in this low detection setting it doesn’t pick up actual movement in the yard by a person or dog. So basically it’s only good at detecting bugs and rain drops.

I really hope this can get addressed on a future firmware release or this cam is only good if I actively open it up to look at it.

1 Like

I am assuming this happens in low light and not during the day time? I have three V3’s and I had this as well with IR on at night. I tried the near and the far IR setting and sensitivity changes to no avail. I finally turned off the night vision IR and now rain and snow and bugs no longer set it off. You don’t have night vision but you don’t wake up to 200 false alarms from snow, rain and bugs.

Just touching base, do you know if there is any fix, or the progress, for Android Phones not having the Tag List to select from when submitting an Event?

1 Like

think of the scale as a “percentage” point at which detection is enabled. bugs and droplets would be close to the camera, thus taking up more of the frame when they zoom by. so it looks in relation to the frame as if it is something large( especially if the cam is outside with the IR on and the bug or droplet reflects the IR light). but something further away relative to the camera appears as a smaller percentage of the frame and thus would not be picked up at those distances with the lower settings to avoid the bugs and droplets. this is one of those instances where it really pays to have cam plus as you can capture everything with the detections settings at max and then filter out what you don’t want to see using the cam plus filters.

4 Likes

can you try to update to the latest doorbell firmware and turn off and on the detection toggle again? we do really see some bugs in early firmware which cause the person detection with other detections not working

For the face, we are looking into the issue now

3 Likes

yes, this is fixed in the latest internal build. will be released soon

3 Likes

This is where the sophistication come in and we are fixing right now. We’re still in the process of internal testing exact the same situation you mentioned. In general, we are not training the model to “guess” the object right now, which is quite difficult to do and requires lots and lots of training data. We value the detection zone in this case, and based on the zone info to detect the object. I’ll share more info once we have more concrete results from our test

5 Likes

Appreciate feedback. Kudos to the team. looking forward to the release for testing.

1 Like

Ok the doorbell is up to date and I disabled face recognition then enabled after 10 seconds, will have to test this week before Monday if this did anything. Maybe the v2 detection will be more accurate for doorbell.