Google coral ai to offset cpu usage


I am using bluechery a while already and i always struggle with cpu usage on motion detection, since my server can not hold a major graphics card…i also dont want to run a 500 watt monster for just checking my chickens…
So i saw that Frigate is doing the calculations on a 60 bucks coral usb dongle…
Which is open source and all…
Can that be integrated as you did the graphics offset for motion?
Thx for any ideas…

This is not exactly 100% related to your question. How ever I also fought the 100% CPU usage. Unless I have missed a later update of the 3.0 DVR, having a GPU card in the DVR server is not going to have much effect. In fact I have my DVR system as a virtual computer in VMware with basic settings for video. I would fight the very high to 100% CPU use all the time and could never get it to a normal 30% or less value. And this is on a dual quad core Xeons with 48 gigs ram in the system. I now just have all 18 cameras set to continuous record. As soon as I set everything to continuous record, I am maybe using at times up to 30% CPU and I never miss any events since it is continuous record. Before when I tried motion detect, I would not get recordings of motion or events that should have been recorded and were not, and because it would peg at 100% CPU use I was not even able to web into it or use the client to access the cameras or events. Might just been me. But you might just want to set your system to continuous record and see how things run. I have 2 TB of storage allocated and running 18 cameras on the system. I have 2.5 to 3 days of retention storage which is as far back as I would ever need to go. I have 15 minute recording sessions, so its not like I have one super big massive recording file.

Again just a thought / idea if you might want to just try running the cameras and DVR that way. Just let all cameras record around the clock. No motion detection. Since I went to continuous record, my system has been very stable, responsive and I have never had a missed event or missed item of concern not recorded.
Best of luck.

you can get 2U gpu’s, that are powerful enough, and lots are offered second hand.

So, I run a box at this time, 4 cams connected, motion detection, and it doesn’t nearly does a thing. Box runs an AMD R5 5600, with 16GB of ram.
No vaapi at the moment.
I miss no recordings, while the machine suffers at the moment due to old, first gen sata hdd.
Triggering was actually to sensitive.
Would love to integrate an extra gpu, hoping we can get some selective object recoginition motion detection., unless I miss that somewhere.