License Plate Tracking Has Gone Open-Source
OpenALPR makes it all so, so easy.
Image: Steve Jurvetson/Flickr
Computer vision-assisted license plate reading seems to be a favorite spectre of a certain sort of privacy worrier, at least anecdotally speaking. Something to do with its potential for tracking and profiling an activity that seems to make Americans in particular feel their most, uh, liberated (driving cars).
The technology has had a steady uptake among law enforcement agencies, who of course think it's wondertool for busting crimes, but now an open-source implementation is starting to make some waves in the developer world and beyond. This is OpenALPR, and, as Mike James notes at I-Programmer, it's not a new thing but is getting some new attention in the wake of a recent code release. LPR is here for the masses and it's incredibly easy to use.
The fundamental task of Openalpr is to take an image file and return text. That's all. It doesn't query government databases for you, and the applications suggested mostly seem to be kind of contrived-sounding small-time security schemes. The promo video is a bit off:
OpenALPR works well and fast, at least judging by the demo. It's also legal for the most part. As EFF lawyer Jennifer Lynch tells Ars Technica, "While a handful of states have passed laws explicitly restricting private citizens and companies from using ALPR technology, outside of those states, there is not much in the law that would prevent someone from using the technology unless its use rises to the level of stalking or harassment. License plates are exposed to public view, and ALPR companies like Vigilant consistently argue they have a First Amendment right to photograph plates and retain the data they collect."
The Vigilante Lynch mentions is Vigilante Solutions, the almost cartoonish dark-side to private LPR tech. Vigilante sells LRP hardware, but goes a step further by maintaining an actual license plate database of some 550 million entries, largely provided by a Texas-based auto repo firm called Digital Recognition Network and largely employed by law enforcement agencies. (Freelance repo men scanning traffic for targets is a pretty creepy idea though.)
On the other hand, this stuff is also pretty neat, from a technology/algorithmic perspective. If you were interested in learning a thing about object-recognition in the real-world, the OpenALPR codebase seems like a decent place to start.