My immediate thought on this is that it's going to be extra-challenging for a few reasons:
1. The ID is normally on the tail (er, vertical stabilizer), which will necessarily be moving perpendicular across the field of the camera, rather than more-or-less directly toward or away from it.
2. Coupled with the higher speed you're likely to see, there's a far greater chance of motion blur obscuring the ID.
3. Add to this, LPR on cars can normally be focused on a pretty small area as cars go through a "choke point"; with planes, there's a WIDE range of heights the ID may live at, meaning a wider camera view, meaning less detail.
4. LPR is generally aided by the fact license plates are made to be highly reflective, with high-contrast letters, which allows the function to be aided by relatively low-power illumination/flash, ie. it takes very little light for the plate itself to be brightly lit, while the letters aren't. Airplane IDs are also high-contrast, but the surfaces aren't generally that reflective, requiring more light (brighter flash), which in itself would be hazardous to other traffic.
5. I would expect range from camera to subject would be much greater than typical with LPR, meaning weather (fog, rain, snow) is going to be a greater impediment than with LPR.
Given all this, I'm not surprised that a *workable* solution would be quite expensive. Keep in mind that true LPR was traditionally quite expensive until just recently as well... takes a while for the economy of scale to kick in.
(Bosch Rep) We did a proof of concept for high speed, low light camera imaging at a small airport in Virginia. While we didn't have the requirement for OCR, I can't imagine that being impossible, but it would likely be the hardest part. Here is what we found during the test.
Not to act like they didn't think of this but aren't most airplanes equipped with some sorta transponder, so that ATC can identify them?
Maybe they are supplementing their ATC database and/or need visual proof to charge landing fees?
As a former pilot, yes planes do carry Transponders. in fact, if you what to see a particular plane in flight right now go to PlaneFinder . Click on any of the aricraft or you can search for a specific aircraft. The "Squawk" number shown when you click is the aircraft's transponder code assigned by ATC.
I assume you wish to find appropriate software to OCR arbitrarily-oriented text from natural scenes, preferably from an H.264 stream.
I Googled "arbitrarily oriented text," "natural scenes," and "video." This buries you in research papers. I examined the authors looking for viable companies that might have a relevant product. In 20 minutes, here's some interesting capabilities I found:
These similar (but not adequate) capabilities suggest that such a thing is possible, and probably available, but we're running into the high cost of information. None of these really fits the bill, and who knows if anything will, but searches are so full that it's a challenge to winnow down to useful results, rather than a sparsity of results.
Another approach (which I didn't take) is to look at moderately recent papers (say, from 2010-2011) that have been cited in the field and contact the authors to ask about commercially available products that work on video. Subject matter experts are more likely to know about the state of the art and available products.
I did a project similar to this for a small regional airport with cameras on the taxiways. It wasn't "recognition" but rather "capture" so that an airport employee could review the recorded video to document which planes took off or landed on the runway after hours for billing purposes.
The major issue with tail number "recognition" would be the varying fonts, colors, locations and sizes of the tail numbers and reflectivity of the background. License plates (even though they vary by state) are fairly uniform in size, reflectivity and displayed location.