OK so can’t comment on latency due to camera. A fair benchmark test would be to compare the time for face detection for dlib/facebox from an image file. On my Mac (8 GB ram, localhost) face detection (recognition OFF) takes about 0.5 sec. In production I run facebox on a synology (also 8 GB ram but less CPU and also over network) detection takes about 2 seconds. @arsaboo is running on a NUC and getting great performance I think.
Of course one major benefit of facebox over dlib is that you can run facebox on ANY computer on your network, whilst dlib must run on the machine hosting HA - which if an RPi is a serious constrint.
Good to know, I’m getting similar face detection times but I’m running Home Assistant on a Xeon E3. I’ll give it a go on the next Home Assistant release, thanks!
Rob. Got it running. Works great!
Im running a NuC clone (AMD A8-5545M) with 8gb RAM.
created an automation to take a picture save it as the local_file camera and announce the recognised picture via tts.
takes about 5 seconds.
@juan11perez use of TTS is cool - are you using recognition so it announces WHO its recognised? I had in mind an automation so Alexa would announce who is approaching my front door (if recognised)
Rob, that’s the intention. It takes the snap on motion trigger and announces via TTS using Chromecast.
I didn’t try alexa since currently you need to ask it.
thanks @robmarkcole this is working perfect on my Hassbian.
guys i have question, can i install facebox and tagbox on the same docker, i have facebox running and wanted to install tagbox but it’s giving me an error about the port 8080, i’m not sure if need to be on a different host, thanks
Rob, so on your optimisation recommendation; I think it’s a must. Overnight it bled about 0.5G of ram on the default scan interval. would this be the correct sintax for calling the scan service?