Top level plate and confidence is not the most confident


I am using the openalpr64-sdk-2.6.101 SDK. Occasionally the top level plate and confidence is not the same as the first candidate. It doesn’t happen very often, but it does happen.

Here is an example:


The default behavior is to pick the best result that matches the plate template for the given region. In this case, the first plate to match the Indiana plate templates was this one:
“plate”: “9397”,
“confidence”: 59.19476,
“matches_template”: true

Dropping your top_n value to something like 5 or 10 would remove a lot of these spurious low confidence reads so that this is less likely to happen. Also, you could ignore this behavior and always pick the top result from the candidates list if you prefer.