absolute beginner advice.

[Robinlawrie]

Hi all, and thank you for your patience if i ask obvious questions, and apologies if this isnt the right place for these questions.

I'm currently doing up an old countryside house, and wish to install a self-hosted security camera setup. It looks like i would need 5-6 cameras to cover all approaches to the house. my desire is to use the cameras to record 24/7 but send me alerts when somebody enters the garden, and possibly trigger a verbal warning or turn on lights.

the most important thing for me is to have the most reliable person detection possible (as it is for everyone i expect) We have a dog and i also dont want to be woken at 3AM for no good reason.

That has lead me into a hugely complex world of TPU's, AI models etc etc, and im having a little difficulty parsing the options.

im an utter nerd, very technically proficient with electronics and have built every workstation ive ever owned (3d animator by trade) however the whole AI thing is something ive never looked into until now. happy to learn, but im hoping to get some answers so i head down the right path at least:

so, here goes:

i see people often use BlueIRIS and Frigate, its not clear why, if blueiris can do the object detection using codeproject AI.. is frigate better for detection?

Ive seen the Coral TPU suggested, however ive also read that it limits the AI model size used, and the images it uses for detection are reduced in resolution. this does not seem a recipe for the best detection accuracy possible. its noted as being fast, but i see people talking about 15 ms vs (for example) 200 ms detection times for slower options.. 200ms seems perfectly fast enough unless i am misunderstanding something.. ?!

an alternative appears to be Openvino on intel integrated GPU.. this (if i have understood well) would allow larger /more varied AI models to be used and higher resolution detection images.. is this the case? would it lead to more reliable detection over coral?

are the two above options available in Blueiris?

is there a current "gold standard" for self-hosted person detection accuracy? (i dont imagine i will ever have time to create a custom training set of labelled images for my specific cameras, even if that seems the ideal method from what i have read.)

as you can see i have lots of questions, and the internet does not seem to want to provide clear answers..

in case it makes a difference, im a windows guy. never used Linux as my work software is windows-only. so ideally id stay in windows to avoid another mountain of new stuff to learn.

any help most appreciated.

[MikeBwca]

...

[HeneryH]

Uh, sorry, this is not an absolute beginner question!

[Robinlawrie]

thanks MikeBwca, and , HenryH, its "absolute beginner" in the sense i have never used or seen any of the software of which i speak, never installed a cctv system, and have only been trying to figure out the best way to protect my eventual new house for a couple of months.

my main concern since the beginning has been to find an overall system that wont cost the earth/have monthly fees, wont sound a 120dB alarm at random times and/or when the dog wanders in from the garden (id like not to make enemies of my new neighbours) whilst at the same time acting as a suitable deterrent to thieves.

that process led me to a decent alarm system covering all points of entry to the house, and a few key points inside with dog-safe sensors. , plus a separate cctv covering all approaches to house, which would warn me/them/the police before they arrive at the house or trigger the alarm. that led me to AI cameras, and from there to BlueIris and pc based AI. the nerd in me says this will suck up lots of my time, probably cost more than expected, and be fun to fiddle with. sold.