OK nerdstr:
How do I take surveillance system video streams and feed them into locally hosted AI to detect and identify different types of wildlife, humans, vehicles, etc?
Discussion
https://opencv.org/ could be useful
I would think you’d need a full-size server rack filled to the brim with top-of-the-line Nvidia GPUs.
I've been looking at this, and ZoneMinder or Frigate seem to be a starting point.
ZoneMinder appears to be ok but not perfect. I've seen a bunch of flowers being mistaken for a cat at a higher confidence than the actual kid as a human.
Unifi has some of this, but not open source afiak.
Frigate NVR https://frigate.video/
LAN only Unifi NVR. Not open sauce but does the job.
Can you grab the stream via USB? Use some model like imagenet and feed it frame by frame
Good evening #nostr
and another job going to AI. Good? Bad? I don't know, but we will find out in a few years.
OpenCV for this as a basis. there is also OpenCV-Python, but you need to know how to use numpy, i used to use this to do image detection experiments faster
I use Frigate.
you can use a HDMI to Ethernet adapter. it takes the video and turn it into a mpeg stream
OpenCV
Since no one mentioned it, yolo might be the easiest AI model to use for this: https://github.com/ultralytics/ultralytics