So just...turn them off.
If you didn't know, detection runs all the time for anything of this sort unless you have what we call "predetection filters".
If you don't have predetection on video, EVERYTHING gets evaluated. So you pare down what is to be detected by certain, faster events.
Motion detection is probably the most basic because it's done very quickly in software, and almost certainly by ffmpeg, costing very little in system resources.
If your machine is struggling with however many feeds you have, do this:
- Reduce the frame rate each camera sends to the minimum acceptable level
- Set image compression to the minimum acceptable level, and whatever your hardware is fastest at (read: not RAW anything)
- Set a motion filter to only run model detection AFTER motion detection
That should sort you out.