I've been working on this for a few months off and on, but now it's refined enough that I at least feel comfortable making a brief announcement. It's a bit of code I've named "osm-makeroads"; basically, it's a semi-automated framework for downloading all the GPS points/traces in a bounding box and using some algorithmic magic to find traces and points that seem to follow the same route and then average them using a statistical technique called principal curves analysis. I can't claim the credit for inventing all this; the "meat" is from an academic paper presented at a conference about 5 years ago:
Chris Brunsdon, 2007, "Path Estimation from GPS Tracks," Proceedings of the 9th International Conference on GeoComputation, National Centre for Geocomputation, National University of Ireland, Maynooth, Eire. http://www.geocomputation.org/2007/1B-Algorithms_and_Architecture1/1B2.pdf The only real complication is sorting out all the overlapping traces, which is what the bulk of the code does before handing each "group" of similar traces off to the principal curves algorithm. The code works best with un-thinned, trackable traces but it'll cope with thinned stuff too; unordered points will also be added to help the fits. Obviously the benefits kick in once 2 or more distinct traces are available of the same route. I've used it quite a bit for remapping (I-77 in NC and SC, I-69 in IN are among the earlier examples) and new mapping (such as the new GA 243 bypass in Gordon, GA, just opened yesterday) alike. It's probably not perfectly robust yet but I'd imagine it should help people who are mapping new areas without imagery, or where there are lots of traces but incomplete local maps. The code is on github at https://github.com/lordsutch/osm-makeroads. It requires R, gpsbabel, and a few R packages (all are on CRAN, some are packaged for Debian, some aren't). So far it's only tested in Linux but the code should run on OS X as-is; might need a few tweaks to the system call code to run on Windows. Chris -- Chris Lawrence <[email protected]> _______________________________________________ Talk-us mailing list [email protected] http://lists.openstreetmap.org/listinfo/talk-us
