I've completed my list of desirements and have declared my Movidius NCS Python code project "finished". Its complete because I'm moving on to OpenVINO and the NCS2 and/or the "new" Google Coral TPU USB accelerator.
I think I've squeezed out about all I can get by using multiple Movidius NCS sticks and/or OpenCV dnn module CPU AI threads. The NCS v1 SDK is a dead-end and the v2 SDK was a waste of time and effort because as far as I could tell it was only different, not better, and didn't support the NCS2.
Node-red provides the ftp-server "front-end" so it can interface with existing security DVRs, MotioneyeOS/motion systems, or stand-alone netcams -- pretty much anything that can ftp a jpeg image to the node-red ftp server (or push a jpeg image buffer via MQTT).
The code can directly fetch jpeg snapshots from Onvif-type cameras using http requests and/or open and grab images from rtsp video streams. Meaning you can mix and match "ftp cameras", Onvif snapshots, and rtsp streams.
I've also provided a sample "controller" flow with a dashboard to enable/disable the system and provide alerts upon detection. Its a starting point for integrating my Python code as an add-on to your security system.
I'd like to put this thread: Standalone Raspberry Pi Security Camera with AI "person detection" to bed and move further discussion here as that thread despite having been a great resource for me, has gotten unwieldy
Questions about my node-red sample flow here please, so real node-red experts can have input and suggestions for improvements.
Python code questions/issues please raise an issue on GitHub and I'll address it as quickly and best I can there.