Is there any way to tweak settings so that AI Tool is more likely to trigger BI? My tolerances are already 0-100% on AI Tool, yet there are plenty of images it scans where a person is in the picture but it fails to see the person and incorrectly calls it a false alert.
I've been messing with Coral USB Accelerator to process every single frame. I've bumped my confidence to > 40%. Sure, it will miss persons in some frames, but in the end it will always have identified it in one of the others... Will share here once it's ready for prime time.
No, because that does not process all frames. My script is working on the RTSP stream.Are you using: robmarkcole/coral-pi-rest-server w/ Pi?
No, because that does not process all frames. My script is working on the RTSP stream.
@GentlePumpkin apparently 5.2.9.15 allows the memo and flagging now
Works for me too now, nice!
Blue Iris motion detection triggers the camera, and as soon as AI Tool confirms the alert, it gets flagged and has the detection tag. Thanks for the heads up
I was one of the initial testers, couldn't get it working too great on Windows. It would work for a while then completely bomb out and stop processing for a few hours. Decided to give docker version of Deepquest a shot - wow! It's running really well now, especially with GentlePumpkin improving the AI Tool.
Using version 1.65, sending alerts to my Alexa devices via notifyme when someone is detected at the front door, no more senseless recording either of car headlights either. Cool!
@GentlePumpkin Face detection could be quite nice in certain instances, such as for automation. Example, turning on the hallway lights automatically when a known member of the family is home. I've got a 4K camera at the front door - how much of a processing delay is there?
I had a question as well - When AI Tool is running as a service, is there a 'proper' way to view stats? I generally like to babysit the tool for a while to gauge how well it works and tweak masks etc to make it work better. The AI itself seems to be a lot more accurate than when I initially tried this out.
Haven't yet figured out why.@GentlePumkin is there any update on the bug where you delete the history,csv but the stats graphs and figures won’t reset?
It's way simpler! I'll explain it in detail later. Meanwhile, you can still use the duplicate camera setup if you prefer that. I always try to update AI Tool in a way so that you can stick with a specific setup if you prefer it. Sometimes it's not possible, but I always invest alot of time trying to.
usually AI Tool automatically deletes entrys for images that were deleted during runtime. And starting takes a few seconds with a large history because AI Tool is checking if all entries are still valid. Your problem though indicates that there might be something wrong with the code that monitors the input folder.
Thanks for the reply! And i am interested in it for all the reasons above as well as a couple more, like for technician recognition for access to restricted tools etc... (i own and operate an auto repair facility that is completely automated), not to mention the cool factor would be through the roof! LOL . I am running my automation system as well as security and surveillance on a dell 720 server with dual zeons and over 100 gb ram so i am a little better off than some in the power area and my 13 cameras are 2 mp 1920x1080 so they are border line. I certainly wouldn't want to cause reliability issues with the current software, it's rock solid here and works flawlessly. With that said I would be glad to beta test if you or anyone were so inclined to make this a feature that a person could switch on if their hardware and situation would allow reliable use. And turn off if not. Thank You again for your valuable time!I tried the face recognition feature. I guess most people would like to use this to get information like "person xy is home" or to block alerts when a known person is detected. Generally a good idea I had in mind too, but I think it would undermine the reliability as p.e. anyone could just print your face and wear it. Additionally and probably most importantly, the face recognition seldomly works if the face resolution is as low as it usually is on p.e. a 2MP camera. On higher res cameras this probably is a less important point. Third point is that face detection is significantly slower than usual object detection in my experience (can someone confirm that?).