choppy during live and AI question

[Blktbss07]

im using a EmpireTech Color4k-B180 (will be connecting more once i figure this out)
i7 processor
32gb ram
windows 11
CPAI on linux docker (runs quicker and better)

So might as well put a couple issues and questions into a single thread.

During live view the video stutters and i guess is ghosting and pixelating. The only issue during playback is its choppy as well, like it skipping a frame.

AI keeps alerting on the same objects over and over, its listing that flower pot as a person, that light as a suitcase and the truck. how can i stop it? I want to use "Ipcam-combined" but the logs are showing that its using all the models.

im sure im overlooking simple fixes.

Thanks!

 

[Spuds]

Without a lengthy response, here are some areas to get you started

1) Uncheck "Use default models" so it will instead use the custom ones (or disable default ones on a camera by camera basis)
2) You currently have 75 real-time images at 200ms AND something in the "to cancel" box. This will result in 15 seconds of AI work as you are forcing it to analyze each those images to ensure that nothing in the "to cancel" was found and to find the "to confirm" image with the highest % confidence. You can reduce that 75 to 5 which will look at a second of video, or simply clear the "to cancel" box and let the system stop looking once it finds a person,car,truck above your 50% mark (but still reduce that to 4-8 images)
3) Your sub stream resolution is high enough that you can uncheck "Use the main stream if available"
4) Enable static object analysis to stop the flower pot etc from being picked up.
5) Under your web server encoder profile, your GOP is set to 25, change that back to 300. You quality is set to 80, set that back to 50. It appears it should be using "direct to wire", but if its not then those settings are way off.
6) Your camera main stream bitrate is set to 4096, that is to low, 8192 is the min I would set for that camera. Change FPS down to 15
7) If that camera has IVS, use that and enable ONVIF as a trigger

 

[Blktbss07]

Without a lengthy response, here are some areas to get you started

1) Uncheck "Use default models" so it will instead use the custom ones (or disable default ones on a camera by camera basis)
2) You currently have 75 real-time images at 200ms AND something in the "to cancel" box. This will result in 15 seconds of AI work as you are forcing it to analyze each those images to ensure that nothing in the "to cancel" was found and to find the "to confirm" image with the highest % confidence. You can reduce that 75 to 5 which will look at a second of video, or simply clear the "to cancel" box and let the system stop looking once it finds a person,car,truck above your 50% mark (but still reduce that to 4-8 images)
3) Your sub stream resolution is high enough that you can uncheck "Use the main stream if available"
4) Enable static object analysis to stop the flower pot etc from being picked up.
5) Under your web server encoder profile, your GOP is set to 25, change that back to 300. You quality is set to 80, set that back to 50. It appears it should be using "direct to wire", but if its not then those settings are way off.
6) Your camera main stream bitrate is set to 4096, that is to low, 8192 is the min I would set for that camera. Change FPS down to 15
7) If that camera has IVS, use that and enable ONVIF as a trigger

I appreciate the reply!

Ive made all those changes and all the issues still occure. Now BI isnt labeling AI detections, and the server is still using all the models.

 

[Spuds]

What does it show under alerts (not all clips) ... I think the burn only goes on the alert image

Also I forgot to mention, when you changed the FPS on the camera to 15, change the i-frame interval to be the same (15)

How do you know its still using all the models? Maybe restart CPAI, I think you said you have it running in a docker container.