IP Camera automation

I’m new to having ip cameras, so here’s what I’m doing so far. Would love to hear more ideas!

  • I have dahua cameras which have smart intrusion / human detection.

  • I have Blueiris recording continuously from the camera’s RTSP stream. It also provides mjpg / hls / jpg snapshots which is extracted from the rtsp stream. This is used for my wall display and google home.

  • Using ipcamera binding, I get alerted when a person is detected within the designated zone.

  • When a person is detected, I would show the mjpg stream of the relevant camera on a 24" wall display using mainui custom widget. At other times, the wall display would show photos from selected Facebook albums.

  • Also when a person is detected, I would take snapshot images, and send them to Deep Stack for facial recognition. Recognised faces will get announced over Google Home, e.g. “Jack and Johnny are on the front porch | driveway”

  • Just for fun, I have items with ga="Camera" tag that contains Blueiris’ HLS url, so I can say “OK Google, show the front porch camera”. Not really sure how useful this is.

I also have a dedicated camera to capture license plates of cars that pass through the road. I haven’t quite done anything with this yet. What’s a good and free way to perform license plate recognition?

What fun and useful things do you do with ip cameras?

How accurate is the facial recognition part?

I’m personally looking into hooking a NVR into OpenHab.

I’m also keen on finding out if there is any kind of access to Law Enforcement Facial Recognition
or similar databases, so that the intercom system can be alerted with a tag “Possible match of wanted fugitive” when said bad man is at the gate ringing the sip video bell. (though based on hollywood movies, technology may not be fast enough to serve realtime requests)

At the moment, I’d say it’s not that accurate since I haven’t tested it against strangers. It detects me and my wife’s faces correctly most of the time and it’s quite fun. Giving it good samples is the key I think, also in comparison to the actual resolution / size of the faces that it can get (depending on the camera framing - how tight the faces are).

Doubt very much that such a resource exists for the general public, and even if it does, it would take a loooong time to run a match. Can you imagine how big of a computer farm they’d need to have to let everyone have a go?

I run 4 snapshots of 1080p images against deep stack, which is hosted on another computer on the LAN, and that took several seconds. That’s with a computer with CUDA GPU, albeit a rather low end one.

1 Like