Does using AI for motion detection reduce minimum system requirements enough to run on an i3 Nuc?

[Marble68]

So I do some AI tinkering using Intel Movidius sticks on the RPI as well as windows development, and saw that someone here made an AI plugin for DeepStackAI.

I found it in this guy's video:
DeepStackAI supports the Intel Movidius neural processor on the Raspberry PI (Alpha). The Intel Movidius can be used in windows just as easily and I see that the project is open source.

So I was wondering - Do you all think I could run Blue Iris + the AI detection on an i3 processor w/ 16gb of ram?

What I would do is fork the current project on GitHub ( VorlonCD/bi-aidetection: Alarm system for Blue Iris based on Artificial Intelligence. (github.com) ), and add native support for the Intel Movidius in windows to the project.

Of note, for me, is I'd be motivated to get the Gen 1 sticks working (would probably have to use a different network, but who cares if it works - YOLO would probably work just fine!) - because the Gen 1 sticks are slower, they're cheaper and easier to find.

My thinking is if all the detection is being done by the AI, the CPU / memory requirements would be lower?

I have a remote cabin with cell based internet access so sending to the cloud isn't an option, so running the AI locally is the way for me.

Does anyone have experience with this? What's your experience?

My specs would be an i3 based Intel NUC, 16GB of RAM, SSD for m.2 for boot, 2TB+ 2.5 inch for storage, and an intel Movidius + 3 foscam 1080p wifi cameras.

This way - I'd only get an alert through my LTE connection if a person was on the property, so my LTE data wouldn't be burned up.

I'd be very thankful for any feedback & suggestions on this.

 

[spammenotinoz]

If you enable sub-streams, direct to disk recording and hardware decode should be absolutely fine.
If it does struggle try H.264 not H.265 much lower CPU requirements.
Deepstack is quite low on resources, I wouldn't even bother with the Intel Movidius, especially under windows just too unreliable.
Use the clone method to ensure "sub-stream" images are being sent to AI Tool\Deepstack, or have BI re-size the Images before passing to AI Tool.
Note though: Using BI alone will use less resources than with AI Tool, so I may be missing something with your post.

 

[SouthernYankee]

A i3 does not support hardware decode. A NUC is not designed to run 24/7/365 under load. Most NUC use laptop processors.

I work recommend buying a used HP/DELL Business computer off of EBAY. An I7 or I5 that is 6th generation or better

Choosing Hardware for Blue Iris

Buy refurbished. Don't build or buy new. The most cost-effective Blue Iris computers are refurbished business-class systems such as "HP EliteDesk" and "Dell OptiPlex". You can usually get a good deal by searching [URL='http://www.ebay.com/']Ebay[/URL] for specific CPU model numbers, and...

ipcamtalk.com

 

[spammenotinoz]

A i3 does not support hardware decode. A NUC is not designed to run 24/7/365 under load. Most NUC use laptop processors.

I work recommend buying a used HP/DELL Business computer off of EBAY. An I7 or I5 that is 6th generation or better

Choosing Hardware for Blue Iris

Buy refurbished. Don't build or buy new. The most cost-effective Blue Iris computers are refurbished business-class systems such as "HP EliteDesk" and "Dell OptiPlex". You can usually get a good deal by searching [URL='http://www.ebay.com/']Ebay[/URL] for specific CPU model numbers, and...

ipcamtalk.com

While you are correct in your recommendation, you are however wrong about i3's not supporting QuickSync. I have a old i3-8109U in a 2018 NUC and that definitely supports quicksync and happily runs a 6 4k cameras.
NUC's are designed to run 24x7, which is why a lot of people buy them, even though they are over priced for what you get, they are very energy efficient.
Before BI had Sub-Stream support, on this device I was happily using 6 cameras configured with motion on the sub-stream to alert the respective partner camera.
It could not however run them without sub-stream support.

My main BI rig is an i7 refurb from ebay, which has a crap ton more grunt. To OP, the processors in the NUC's are the "Low Power Mobile" -U editions which have much less grunt than the non-U versions.

Actually: I gave my first NUC away to a friend who will use it as a Plex sever, I had it running 24x7 since 2015. It had a noisy fan, ordered a spare off ebay and replaced it. Hope he gets a few more years out of it. Although under powered and over priced, I am a fan of the NUC.

 

[Marble68]

Ok good info.

Thanks to all for the feedback. This gives Me some direction and ideals.

Why is Movidius considered unreliable on windows? I’ve had zero issue with them on PIs for CV.

 

[Marble68]

A i3 does not support hardware decode. A NUC is not designed to run 24/7/365 under load. Most NUC use laptop processors.

I work recommend buying a used HP/DELL Business computer off of EBAY. An I7 or I5 that is 6th generation or better

Choosing Hardware for Blue Iris

Buy refurbished. Don't build or buy new. The most cost-effective Blue Iris computers are refurbished business-class systems such as "HP EliteDesk" and "Dell OptiPlex". You can usually get a good deal by searching [URL='http://www.ebay.com/']Ebay[/URL] for specific CPU model numbers, and...

ipcamtalk.com

FWIW, I have uses a NUC as a build / dev database server for years with zero issue, and it runs 24/7.

CPU isn’t taxed al the time, so we may be talking apples and oranges.

the only issue I’m having with it after 4 years is the CPU fan buzzes occasionally and I need to replace it soon.