Intersection clustering #59
Conversation
…ntersections. Highways with a common name but different IDs occur when a boulevard is represented as two separate ways. Clustered intersections occur when boulevards intersect. This results in a single "intersection," as a human would consider it, being represented by several way crossings. This is helpful for routing purposes, but not for user interactions. Signed-off-by: Ted Steiner <tsteiner2@gmail.com>
I was getting errors on OS X. Signed-off-by: Ted Steiner <tsteiner2@gmail.com>
Now when a highway is reduced to 0 length, it is deleted in the replaceHighwayNodes! function. Now skip intersection consolidation when intersection has only one associated highway (ends of roads). This prevents the end of a boulevard (such as when cropped) from being joined together. Removed unused clustered_highways variable. Issue #31. Signed-off-by: Ted Steiner <tsteiner2@gmail.com>
…IntersectionConsolidation Pulling in minor bug fix for plotMap. # Please enter a commit message to explain why this merge is necessary, # especially if it merges an updated upstream into a topic branch. # # Lines starting with '#' will be ignored, and an empty message aborts # the commit.
Now if a highway has been removed from highways but still exists in the roadways classifier, it will simply be skipped during highway segmentation. Previously this resulted in a key not found error. Signed-off-by: Ted Steiner <tsteiner2@gmail.com>
Changed type HighwayCluster to HighwaySet. Intersection clustering requires nodes to be in ENU coordinates. Issue #31. Signed-off-by: Ted Steiner <tsteiner2@gmail.com>
…documentation. The list of cluster IDs is no longer returned, since it can easily can be extracted from the intersection ID to cluster ID mapping dictionary. Issue #31 Signed-off-by: Ted Steiner <tsteiner2@gmail.com>
Signed-off-by: Ted Steiner <tsteiner2@gmail.com>
|
Very cool. It will be interesting to see how this plays with @yeesian's work on turn restrictions / relations -- exciting to see a lot in motion! |
|
Yeah, I'm not sure how it will go with turn restrictions. Probably we will need to make the function that replaces nodes in the highway list smarter, to make sure it preserves the turn restrictions, shifting them as necessary. In general use, it's probably not necessary to have them both working together. If you want accurate routing, you'll probably use the full graph. This is more about counting the intersections. And it would also be useful for "human readable" driving directions, so we could give a list of intersections. Even in that case, though, all that would be needed is the mapping between true intersections and their associated clusters (the output of the clustering function). |
|
It looks like Travis failed only on the julianightlies branch, involving reading the xml files with libexpat: Has anyone tried running `Pkg.test("OpenStreetMap") with version 0.4 yet? |
|
Yep, I generally work on 0.4, and then test at the end that I'm also 0.3 compatible. Merging #55 passed Travis on 0.4, and I also tested it locally on 0.4. I've tagged a new version of LibExpat.jl since then, though -- it was tested alone and under OpenStreetMap.jl on 0.4 before tagging, but who knows. I can't look more now, but I will try to look later. |
|
I tested this branch on 0.4 locally, all passing. Opening another issue for the recent Travis-only 0.4 failures, to handle separately if it doesn't go away in a day or so. |
|
Ok, I'll go ahead and merge it then, despite Travis being cranky. It looks like the master branch is having the same issue right now, anyways. |
This code addition (no changes to existing functions were needed) allows clustering of intersection nodes into a single node. This is helpful in the case of boulevards and divided highways crossing one another and inducing 3+ intersections, rather than a single highway intersection. This allows you to get a better estimate of how many highway intersections are present in a region, because the divided highways will not inflate the count.
There is also a function included to replace all clustered intersections' node IDs in the highway list, so that the highways all pass through a single point. This might speed up routing a little bit in some cases, and also enables findIntersections() to be run again on the new set of highways to get the more accurate count.
I also added tests and documentation for the features.
Closes Issue #31.