Feature: API-Endpoint, Filter Enhanced, Added Features, etc.

[MikeSchrijver]

First of all, I'm not a native Python developer I learned Python while doing Advent of Code and working on this project.

Some context I did some research about open-source/free ANPR systems and commercial ANPR systems. Long story short we choose this system, because it met all the requirements and I learn something on the way.

During the research, I did some testing on how good the detection and OCR algorithms are. Results are as following:

Dutch License Plates: 2/10
Foreign License Plates: 2/10
Time Per License Plate: More than 50 sec*

*: did not have a GPU available, so don't know possible time enhancement of a GPU

So I started working on this system and implemented some new features, enhanced old ones, and did some fixes.

New Features:

1. API-Endpoint (Default TensorFlow Lite**)
2. License Plate Recognizer has now a help text
3. Debug Modus for License Plate Recognizer
4. Print Time Execution and Modal Load
5. Add option to single OCR a character or the total license plate
6. Add option to use Deskew and rotate

**: You don't want the API to be slow so I choose to use Tensor Flow Lite

Enhanced:

1. Filter to be more accurate

Fixed:

1. When running --plate the whole image is used, not the cut-out license plate

When running the same tests the results are as following:

Dutch License Plates: 8/10
Foreign License Plates: 3/10 ***
Time Per License Plate: More than 3 to 4 sec

***: Mostly missed on 0's and O's and concluded that that test dataset wasn't representative of normal circumstances.