New Person Detection Feedback

and if you are you must turn it off to use person detection currently.

3 Likes

Hi. Do all Wyze V2 Cam cameras have a “floating” image? I also have a “floating” image on my Wyze V2 Cam camera. On the other Wyze Pan Cam - no. Thank you.

what do you mean by “floating image”?

could you post a photo of what you are seeing?

what type of environment do you have you camera set up in?

Please, look at video: New Person Detection Feedback - #2 by rbruceporter

Don’t you notice periodic artifacts throughout the video area? As if the internet connection is not stable …

This is especially noticeable in the porch area:

In addition, these artifacts are recognized as movement!

Ahhh ok. I see what you mean there. those are actually called compression artifacts and are due to the compression of the video to fit the given disk space or bandwidth. in viewing this again this looks like its at the smaller of the three compression video settings. when I have limited bandwidth from the connection, the video seems be jumpy. maybe this is recording in SD and not HD? I don’t believe I have these on my cameras, at least not that i’ve noticed. and I have my setting in HD. I will definitely be looking for this when I get home.

the fact that they are being labeled as person has to do with the training method being used for the AI. when starting out with labels such as person detection most motion will be tagged as person and through numerous correct samples sent to them through the app they build their training pool and the AI gets better. currently any light changes I see on my cameras at night is tagged as a person. I’ve sent in MANY videos thus far. When the other company and their detection was first rolled out for beta we had this same issue. and that was especially interesting because they didn’t just have one tag. there was also a vehicle tag they wanted sample feedback for.

one of the benefits to Wyze doing cloud AI for person detection is that they will be able to update the learning pools with samples and improvements far more frequently and without issue hopefully than they would doing so with firmware updates.

1 Like

That’s true. I never thought of that. So the person detection algorithm will improve without firmware updates.

that is theoretically possible. i haven’t heard anything official yet, but it is how it could work assuming there are no huge changes needed to the interface or the connection. the other company did all the detection on the “edge” so everything was being looked at on the camera itself. the downside to that is that if you choose not to upgrade from firmware containing that programming you still have it.

if for some reason Wyze decided to get rid off person detection down the road, by removing it in the cloud it would be gone. so there is a vast difference in control too.

so hopefully this leads to a far easier and streamlined service.

Note that with my Wyze Pan Cam cameras there are no such artifacts. Such artifacts are only present in broadcast and recorded video from Wyze Cam v2 camera.

hmmm and what was the compression setting on for the V2?

also what firmware is it using? if your pan cam is fine, it sound like the cameras itself. so possibly a setting.

Artifacts both in mode SD and HD. Firmware version 4.9.5.98 As far as I understand, this is true in all V 2 cameras?

I would take your camera and put it in another spot and see if the artifacts are still as strong. like I said I haven’t noticed it in mine.

also look at what mediums the footage is pulled from. are you viewing it on the app mainly ( i think someone mentioned once that it is more compressed there)

is the footage from the cloud clips or the sd card. I believe the sd card might have better clarity having slightly less compression. and with that, if it is from the card, fully downloaded to your device using the app or was the card pulled and it copied directly to a computer?

and of course, what sd card was it originally written to. there are huge differences in cards.

1 Like

I noticed that there are artifacts on every street video:

  • as if the image is constantly shifting by a few pixels …

Sorry, I can’t get a link to this video from the Discover section.

Here’s another example sosuch a video with artifacts: Screenshot_20200224-194218

and:

1 Like

check out the captured on wyze section of the forum. im seeing quite a few videos without artifacts, but I noticed most of them are very well lit. I wonder if that has something to do with it. or again, possibly how the video is being uploaded

1 Like

these “twitch” are clearly visible in your video: Tree falls during high winds

@minionsweb, It just showed up the other day on my app. It’s working very well. A few false detections but better than before.

I’ve noticed that person detection does not appear to work if the camera is mounted upside down, or is pointed at a downward angle.

I have 2 cameras mounted upside down, and one on a high shelf pointed down into a room, and none of the 3 are detecting people when they show up on camera. I have another camera pointed out at an angle parallel to the ground, and it does detect people (along with some false alerts for shadows, etc).

I have zero person detection on all cams - Even did a few jumping jacks in front of cam - detected nothing, no notification, I do have “All Other Motion” switched off.

Person detection isn’t working at all on the one camera I have mounted outdoors, which is pointed at a slight downward angle. I can go outside and dance in front of the camera (definitely not something you want to see!) and all I get is a motion event. In my garage, however, I’m getting numerous detections that are nothing but sunlight and shadow.

It’s early days, of course, but at this stage for me at least this is a massive step backward from the xnor.ai detection.

Person detection is not detecting people in any of my cameras. Counts them as regular motion. CMC is not enabled.

So far I’m getting about half or a third of actual person events detected inside, and none on my camera outside. No false positives yet.