The task I am going to solve is to deploy video surveillance system with object recognition and classification connected to HA. Components I need are:
4 dome 1080p IP PoE cameras
2 bullet 1080p IP PoE cameras
Controlling system
RPI 3B+ has been selected as controlling system. Having fed one 4K stream, I realized that RPI3 B+ merely stuck processing even one 4K stream when I need at least six. Am I right that I do not have enough computing power?
If yes, what to choose? Has anyone had an experience in deploying HA to Synology NAS in docker, for instance?
Yes. Very, very right. Video processing is computationally expensive
I’m using Shinobi with five cameras. Four of them are 1080 and the other is 720. The system is only doing basic motion detection. My server is an Intel i5-4440, a quad-core CPU @ 3.1GHz, a few years old now but still very decent for home server duties. Shinobi’s video processing uses around 110% of one core, all the time.
Stop talking about Pis and NASes, and buy a real computer. While you’re at it, pick up a cheap NVidia video card. ffmpeg and other tools can offload some of the video processing to the GPU to save a fair bit of overhead.
Thank you for the input. Generally, you are absolutely right. In view of the fact that most computing power consuming application is video processing, I am thinking over NAS only for video processing and RPI3B+ for the rest.
AI can be deployed by neural accelerators.
Having computer for home automation is really nice, but I am worried about power consumption since controlling system must work 24/7.