Recognize vehicle with no plate


If I understand the pipeline correctly, one of the first steps is to locate plate candidates in the input image. Is the training done in such a way that you could still get a result in case the plate itself is blank or missing? I am thinking of cases where the sun is being reflected directly behind the camera, overexposing the plate, or the plate being covered in snow. Or simply removed.

Is there a way to still run the vehicle classification algorithm even if the plate is unreadable? I realize it may be difficult to group the results into a single detection without any plate information.


It’s possible to disable the portion that recognizes the text on the plate, which would provide you with all possible plate candidates in the image. Add this to /etc/openalpr/openalpr.conf:
skip_recognition = 1

However, you would not get the plate numbers, and there would be more false positives in the output.

Finding the vehicles even when there is no plate is a very common request from customers. It’s a tricky problem to solve, but we are on the path towards doing that, and should have a great feature next year.


+1 feature request. Some no-plate vehicles (neither front nor rear) retain a frame, even when tape used to obscure number, which is possibly of use to an algorithm. Much more common is no front plate, required, but not rigorously enforced in California.