Page 1 of 1
Built new PC, now won't detect people
Posted: Thu Aug 03, 2023 6:14 am
by crespoh69
Hey guys, so I built a dedicated PC to use as a BI and Plex Server and with the upgrade I actually installed the latest version of BI instead of what was previosuly installed. Today I went into my views as I got time to go and tinker with settings here and there and noticed that while it's detecting cars and flagging them, all of my person detections are being dismissed. Even high certainty ones that are 77% sure and the subject is walking slow enough. Did anything recently change in recent versions? Here's an example:
- 2023-08-02 23_13_36-Remote Desktop.jpg (88.92 KiB) Viewed 1923 times
Re: Built new PC, now won't detect people
Posted: Thu Aug 03, 2023 11:05 pm
by crespoh69
Yeah, not sure what the change was. I still had my old drive from the old PC and saw that it was on v5.7.6.8 so I reverted back and it's flagging people again. I'll start updating through the iterations until it stops working again.
Re: Built new PC, now won't detect people
Posted: Fri Aug 04, 2023 10:44 am
by TimG
I can tell you that you are using the default model "Objects" at the same time as one of MikeLuds custom models "ipcam-combined". That will use excessive cpu and gain nothing.
Also does that camera get a close enough image to resolve faces ? If not you may need to disable that too and save even more cpu cycles. Static ? That is to prevent stationary cars etc triggering an Alert.
I think you may need to sort out basic issues like that before CPAI will work correctly. All of the latest software is working well on my home system.
Re: Built new PC, now won't detect people
Posted: Fri Aug 04, 2023 9:54 pm
by crespoh69
TimG wrote: ↑Fri Aug 04, 2023 10:44 am
I can tell you that you are using the default model "Objects" at the same time as one of MikeLuds custom models "ipcam-combined". That will use excessive cpu and gain nothing.
Also does that camera get a close enough image to resolve faces ? If not you may need to disable that too and save even more cpu cycles. Static ? That is to prevent stationary cars etc triggering an Alert.
How would I turn those other ones off? I do need static unfortunately, even in the backyard because I do sometimes work on my car and park it there.
TimG wrote: ↑Fri Aug 04, 2023 10:44 am
All of the latest software is working well on my home system.
I don't doubt it's working on your system but I think it may have run into issues going from an older version to the newest version. On the older version it appears to be working fine now, haven't moved it over just yet.
Re: Built new PC, now won't detect people
Posted: Sat Aug 05, 2023 7:46 am
by TimG
OK, so if you are using a custom model, are you using it on all cameras, or just some ? It is clear from your DAT file that the "ipcam-combined" model is almost twice as fast as the default "Objects" model for you, so I would turn the default model OFF and use the custom model on all cameras. You can turn custom models OFF on individual cameras, and you can turn OFF the default Objects model in general settings. The DAT file is always the key info, as it shows you what is being used on that individual camera.
1. General settings/AI - Untick "Default object detection". When you next have a trigger, you should see in the DAT file that "Objects" is no longer there, but that the custom model "ipcam-combined" is still working.
2. Camera settings/ Trigger/ Artificial Intelligence - To turn OFF models that you don't need on a camera (wastes a lot of cpu) you use :0
My drive camera settings look like this:
- custom.png (12.93 KiB) Viewed 1856 times
So I have Faces and ALPR working on two other cameras, and then I noticed in the DAT files for other cameras that they were working on ALL cameras. I had to go to all other cameras to turn them OFF with the addition of :0 and then check the DAT file for that camera to check what was running.
Early experimentation with DeepStack showed that the :0 had to be used before the working custom model. I don't know if it is still true, but it does work
It is possible that running the two models simultaneously - which are both able to output "person" - causes your issue with certain versions of BI5.