License Plate Recognition from any camera angle

I have an enterprise project which needs to recognize Singapore vehicle plate number, single-line and multi-line vehicle plate. to give you an idea, this will eventually be a surveillance system and will be installed in Jetson Nano, It needs to be able to read above 5 FPS, very good accuracy, and should be able to read from any camera angle, I only need the plate recognition part.

but I have a few questions in mind, please post your proposal with answers to the questions below.

- have you worked with jetson nano before?
- by using your solution would it able to reach above 5fps in jetson nano with a very good accuracy?
- what framework will you use for this project?

I'm a developer myself, I've coded an LPR system using mxnet, but the thing is, it read the plate below 5fps and is not very accurate when it comes to blurred frames,
I've seen some good LPR software able to read blurry images

here's a site to the third party software.
https://www.doubango.org/webapps/alpr/

this is a good example of a result.

cctv video quality example:
https://drive.google.com/file/d/1453pIVCfvW8LZJGHJDHMWd9aDDeM3tu7/view