We know what can be written on license plates, and we know that license plates don't change over short times, so it's possible to determine which characters best match the sequence of frames even if no individual frame is readable. See: https://en.wikipedia.org/wiki/Super-resolution_imaging
