Interesting project! I have a UnitV2 (and the older UnitV) as well but only little experience with. My approach would be the following:
To recognise only a few well known vehicles you can classify the image of the whole plate including car front (and drivers face) instead of the individual numbers/letters on it. This can be done with the V-training or with the online classifier function.
For recognising single numbers/letters you can try to classify individual numbers/letters and then "read" the x-coordinates of the returned boundary boxes in ascending order to convert into string and then compare to a list/database.
