Possible lost matches

[echeipesh]

There is a logical mistake in the match algorithm:

1. First possible matches are computed using straight intersection
2. OSM buildings containing Bing building centroid are considered matched
3. OSM buildings overlapping by 75% of area with Bing building are matched
4. OSM buildings failing that check are dropped from output set

Relevant bloc:

geotrellis-osm-diff-demo/src/main/scala/BuildingsDiff.scala

Lines 104 to 167 in 1ab83e2

	case (Some(msftBuildings), Some(osmBuildings)) => {
	val msftSet = msftBuildings.toSet
	val osmSet = osmBuildings.toSet
	val osmPossibleMatches = scala.collection.mutable.Set[FeatureWithId]()
	val msftPossibleMatches = scala.collection.mutable.Set[FeatureWithId]()
	// First limit comparisons to only geometries that could possibly match
	msftSet.foreach { msftBuilding =>
	osmSet.foreach { osmBuilding =>
	if (msftBuilding.geom.intersects(osmBuilding.geom)) {
	osmPossibleMatches += osmBuilding
	msftPossibleMatches += msftBuilding
	}
	}
	}
	// Construct feature sets where there are no matches
	val osmNoMatches = (osmSet -- osmPossibleMatches).toList
	val msftNoMatches = (msftSet -- msftPossibleMatches).toList
	// Loop osm buildings and check if any bing footprint contains the osm centroid. If so, a match!
	// Note that this algorithm implies a one to many relationship for bing to osm buildings. One
	// bing building can match many osm buildings. Which is fine because in most cases the bing
	// satellite detected polygon is larger and less precise than the actual OSM building.
	val msftPossibleMatchesList = msftPossibleMatches.toList
	val matches = new ListBuffer[FeatureWithId]()
	osmPossibleMatches.foreach { osmFeature =>
	val isMatched = msftPossibleMatchesList.exists { msftFeature =>
	msftFeature.geom.jtsGeom.contains(osmFeature.geom.jtsGeom.getCentroid)
	}
	if (isMatched) {
	matches += FeatureWithId(osmFeature.id, osmFeature.geom, osmFeature.data ++ Map("source" -> "both"))
	}
	}
	// Remove successful matches in last step from consideration
	osmPossibleMatches --= matches
	// Attempt 1 to 1 area intersection ratio check with remaining geometries
	osmPossibleMatches.foreach { osmFeature =>
	val isMatched = msftPossibleMatchesList.exists { msftFeature =>
	(msftFeature.geom.as[Polygon], osmFeature.geom.as[Polygon]) match {
	case (Some(msftPoly), Some(osmPoly)) => {
	if (msftPoly.isValid && osmPoly.isValid) {
	val AREA_THRESHOLD = 0.75
	msftFeature.geom.intersection(osmFeature.geom) match {
	case PolygonResult(p) => if (p.area / msftPoly.area > AREA_THRESHOLD) true else false
	case MultiPolygonResult(p) => if (p.area / msftPoly.area > AREA_THRESHOLD) true else false
	case _ => false
	}
	} else {
	false
	}
	}
	case _ => false
	}
	}
	if (isMatched) {
	matches += FeatureWithId(osmFeature.id, osmFeature.geom, osmFeature.data ++ Map("source" -> "both"))
	}
	}
	// Return all output plus any remaining unmatched OSM geometries
	msftNoMatches ++ osmNoMatches ++ matches

This if needs an else:

geotrellis-osm-diff-demo/src/main/scala/BuildingsDiff.scala

Lines 161 to 163 in 1ab83e2

	if (isMatched) {
	matches += FeatureWithId(osmFeature.id, osmFeature.geom, osmFeature.data ++ Map("source" -> "both"))
	}