Keeps alerting me to empty boxes of ‘person’ detected with 90+% confidence. So frustrated right now.
I’ve been using Frigate for years. The built in object detection (without subscription) is excellent and very near 100% accurate. Initial setup was somewhat of challenge though. It’s free (donations encouraged) or a low cost subscription for more advanced detection.
Every year or so I try to go to Frigate from Blue Iris so I can get rid of my last Windows box. But functionally they aren’t in the same league. Just the PTZ controls on Frigate drive me back to BI within minutes, besides all the rest of the features.
Some day…
Have you had a look at AgentDVR recently?
OK, that looks pretty rudimentary. No MQTT that I could find, no clue how or if it does object recog. Not very serious.
I’d forgotten about that one. I’m going to try it again right now. Thank you.
Any pointers on object detection with that?
I’m ditching BI after years of use after they recently updated the iPhone app which broke me connecting remotely.
And yes, the AI has gone way downhill with each update, be it DeepStack/CPAI/integrated.
I just picked up a Ubiquiti fiber cloud gateway and a camera to start moving over to that ecosystem instead, as my camera feeds are important to me when not at home.
I ran BI for years, but left them for a Ubiquiti device. Yes it required new cameras, and of course the investment in a UniFi device, but damn is UniFi Protect good. Does great detection and no subscription or fees of any kind, plus gets frequent updates and feature improvents. Not a solution for everyone, but I believed the hype of BI for years and thought there was nothing better. Turns out there is.
Try running a Deepstack container in docker and point the AI feature at that container. It’s much better IME.
lol that was the previous set up. I thought since their AI is integrated now, id speed things up and go from the long delays to something more instantaneous. Now it’s so bad with the false positives I don’t even bother checking alerts anymore. I need to go back again.
I’m not super familiar with the integrated one, but I think you can adjust confidence levels in at least one place. That might improve that. Or go back to Deepstack.
Funny thing is the false positives are always 80-90+% confidence.
I don’t know the technology behind it, but back in the day, Google’s Picasa was pretty good at facial recognition and grouping.
