I enabled GPU acceleration and re-ran OpenALPR with the same input, and it ran in 2 hours flat. Still not quite there. I couldn’t get their OpenCL backend working but I assume it would be pretty similar in terms of performance. To my knowledge, none of the free versions enable any multi-frame parallelism, where it seems like the real performance gains might be.
If you hunt down videos of the deployed BlueNet vehicles, they seem to use at least four cameras, which would need to be processed fast enough to be useful.
Part of the point of this article, though, is that just pulling OpenALPR “off the shelf” and running it might not give you “good enough” results in many cases. Would it do so after spending time and money tweaking it, building your own custom training data, etc., or would that just be sunk time and money?