Hey, That's great! I can stop reading about Harris operators. I totally agree about orthogonal snapping.
How well would this scale up to the whole country? (!! Not automatically importing the results of course !!) I'm thinking about tile/batch sizes, tile boundary issues, any necessity for porting parts to c++ for speed etc. robert. On Friday 09 April 2010, TimSC wrote: > Hi again, > > I have been working on auto tracing buildings and I'm making progress. I > was slightly encouraged by Ed Avis's comments. I think one underlying > difference is peoples attitude to omissions in map data. Many people > think they are a good thing, particularly since they encourage the > community to do high quality surveying. But I think omissions are bad in > terms of actually using the map. I don't think we should be using the > main map to gauge our progress. I suspect what we need is good meta data > - how and when data is sourced. Anyway enough rambling... > > Tracing buildings. I have been using the original images, since image > transformations tend to introduce degradation of quality. I use colour > to select building pixels, then form edge fragments, then form polygons, > then simplify the polygons using the Douglas-Peucker algorithm, then > group them so we get inner and outer edges, then tranform image > coordinates to GBOS then to WGS84 via OSTN02 (I ported the perl code to > python), then save as OSM format and load back into JOSM. Screenshot: > http://timsc.dev.openstreetmap.org/dev/josm-building-outlines.png > > The next steps are to improve the quality of the polygon shapes, > possibly by checking if the edges are nearly orthogonal, and if so > making them completely orthogonal. Also I need to write a filter to > check for buildings in the area, to avoid importing duplicate buildings. > I need to look at the simplification, as sometimes an extra node is > added to a polygon (the initial node used as the start of the > algorithm). I am also considering detecting roads that overlap buildings > in the source images, since this is probably the biggest loss of > quality. The result I am getting is already more spatially detailed than > my own survey of the University of Surrey campus (although not as rich > in information). > > In the medium term, I will import some buildings once I have the quality > I want. I want to minimise manual work in JOSM but I don't rule it out. > I will only be working in the Guildford area - it's my data to gamble > around there :) > > TimSC > > > _______________________________________________ > Talk-GB mailing list > Talk-GB@openstreetmap.org > http://lists.openstreetmap.org/listinfo/talk-gb > _______________________________________________ Talk-GB mailing list Talk-GB@openstreetmap.org http://lists.openstreetmap.org/listinfo/talk-gb