AI detection linked to vehicle motion, or just any vehicle?

Thank you for clarifying that the AI algorithm is functioning as designed with respect to these objects, inanimate as they may be in the scenarios I gave. Also, thanks for updating me on the fact that there is a wish list item to have the AI only tag moving recognized objects. I DO understand the difficulties in filtering out superfluous motion that is causing what I deem to be unnecessary notifications. The goal is to allow the superfluous motion to continue to process as motion WITHOUT sending the notifications when only certain AI MOTION tags are in place, e.g. don’t notify me when a car is just sitting in my driveway, but DO notify me when it moves. That I would like to see if it at all possible.

Rewind to before AI tags were available on Wyze, and one DID want notifications for certain movements. At that time, I had to manually filter my detection zone on my doorbell cam so that trees, bushes and my wife’s flower boxes on our front porch did not send me a notification when there was movement sensed. I had to change the filtering at least several times/week. Changing sunlight shadows caused by our wind spinner also created this same scenario. With the advent of AI tags, I now do not have to mask out any of the detection zone, as passing cars trigger a notification but the trees, bushes and plants do not! Changes in light (such as the changing sunlight shadows caused by our wind spinner) did the same, as well, which of course they do not now…ALL EXCEPT FOR THE FACT THAT I DO GET NOTIFICATIONS WHEN THESE MOVEMENTS OCCUR IF THERE IS A PARKED CARD IN MY DRIVE OR A PACKAGE ON MY FRONT STEP! I’m sure that you understand I am aware of the current programming challenges to rectify the problem. Bottom Line: Inanimate objects should NOT trigger notifications!!

On another motion notification problem, I TWICE within two weeks have been called by Noonlight when away from home with my security system engaged. I figured it out in retrospect: A v2 motion sensor sits on my fireplace mantel and there is a pan cam in the room as well. Apparently sunlight from a front window in the early winter afternoons (when we have sunlight this time of year, which isn’t often, LOL) shone on a narrow area wherein the sensor was located. Well the sun only shone on that spot for a few seconds, but when the sun evidently went behind a cloud the change in light on the sensor was enough to trigger motion and set off my alarm–thank goodness Noonlight was able to contact me first before they called the cavalry, as I told them it probably was a false positive which shortly thereafter I confirmed by reviewing cam footage from that room. I moved the sensor when I got home to a spot on the mantel that appears is not in line with any sun streaming in from our front windows. So this, too, hopefully will be looked at by the AI team (if you tell them) with the hope of massaging that respective algorithm?

I thank you for contacting me and hope that the AI wizards can fine tune the algorithms in question, so please pass these observations on to them. I look forward to Wyze’s continued improvements on what they currently offer and to future innovations.

…a further note. Not only will AI flag a motionless car in my driveway every minute with a vehicle notification from my Video Doorbell camera, but even does so with a car parked across the street. These types of notifications are useless, as again I DO want notifications for vehicles in motion In my driveway as well as passing in the street! Why your algorithm for people works well detecting them in motion (as I have tested this by walking towards my front door, then standing still–no falsing), I do not understand why motionless vehicles do send notifications! Should the same sort of algorithm be applied to both humans and vehicles with respect to motion!!! I hope that this can be resolved, because it is a real pain in the as-!

The detected motion of the person moving is what causes the event clip to be saved and uploaded, then the AI looks over the full frame and sees the person and tags as such.

They dont. There is some motion somewhere in frame to cause the event clip to be saved and uploaded. It is then analyzed by the AI and since the car or cars are visible in frame, weither they are moving or not will be tagged.

1 Like

Thank you for clarifying just how AI works in Wyze!

To follow-up, in both person and vehicle recognition, ANY motion in the frame triggers AI to analyze the movement, correct? So if a person was standing still and the swaying of tree branches in the background triggered a notification, I assume that AI would analyze this movement (or any movement in the frame) and send a person notification, correct? Furthermore, if a car was parked in my driveway (or even across the street) and the swaying of tree branches in the background triggered a notification, I assume that AI would analyze this movement (or any movement in the frame) and send a vehicle notification, still correct? In other words, it is NOT NECESSARILY the person or the vehicle in motion, but rather ANY motion in the frame that triggers AI to analyze. I then can assume that the AI algorithm analyzes sizes, shapes, and where in the frame the motion occurs in order to determine if it is a person, pet, vehicle, or package. With this logic, then, if I placed a mannequin in my driveway and the swaying of tree branches in the background triggered a notification, I assume that AI would analyze this movement (or any movement in the frame) and send a person notification, correct? In other words, the triggering of events would be along the lines of those for a vehicle, except the shape algorithm identifies it as a person. Am I on the right track??!!

I am not a programmer or an AI engineer. However, I do question why an algorithm can’t be built that would first analyze the motion of certain sizes, shapes, and where in the frame the motion occurs, identify it as a person, pet, vehicle, or package, and THEN send the notification, having a very good indication from the AI analysis what was the motion trigger. So rather than have random motion trigger AI to investigate, AI constantly would be observing and analyzing, and based upon its findings send a notification if it senses one of the notification triggers.

Now, of course, I do not know if this can be done, and if it can how much computing power it would take and if the system even could handle it. This would resolve the unnecessary triggers that I get under such circumstances. So would you kindly run this by the engineers to see if this is a possibility?

Much appreciated, and thank you in advance.

1 Like

Person, vehicle, pet, package, any of the recognized objects that come with the AI analysis. The camera doesn’t know what it is seeing, it only sees “pixel change” as motion, wether that’s color change, lightness, darkness, any reason a pixel would change. When it detects motion that passes all your settings, the camera uploads the clip to the cloud. Once it is uploaded, the cloud analyses the full frame of the video looking for recognized objects then tags the video with what it detects, then you are notified of such.

Correct, that is how it works currently. This is why you are getting the stationary vehicle tags, because they are in frame.

I am guessing it will tag as a person. You may have to come up with a cool story about what’s happening for when the neighbors start asking what is going on with the mannequin in the driveway. :slight_smile:

Here are some applicable wishlist items you should vote for if you want to help them along!

1 Like