5

I'd like to merge linestrings which touch, but do not cross. The linestrings share the same SRID and come from a PostgreSQL 9.2/PostGIS 1.5 database. Within the table I query are several line strings, not just those I want to merge, but also others.

For example, consider lines from the OpenStreetMap database. I start with a temporary table that has all linestring which are tagged as being part of the power grid, together with the respective voltages. However, not all parts of the grid that belong together are also represented as a single (multi-) linestring, they are within the database the way they were inserted. So, I need to find out which lines touch, do not cross, and share the same voltages, and finally merge them. The result will be another, shorter list containing merged lines and those which did not need to get merged with others.

I've come up with an algorithm in C++ (Qt), which works, but is slow:

for (QHash<qlonglong, MapItem *>::iterator current = gridItems.begin();
        current != gridItems.end(); current++) {
    for (QHash<qlonglong, MapItem *>::iterator other = current;
            other != gridItems.end(); other++) {
        if (other.value() == current.value()) {
            continue;
        }

        const OGRGeometry *firstGeom = current.value()->geometry();
        const OGRGeometry *itGeom = other.value()->geometry();

        if (current.value()->tag("voltages") == other.value()->tag("voltages")
                && firstGeom->Touches(itGeom)
                && !firstGeom->Crosses(itGeom)) {

            // Merge:

            MapItem *mergedItem = new MapItem(firstGeom->Union(itGeom));
            mergedItem->tags(current.value()->tags());

            *current    = mergedItem;
            *other      = mergedItem;
        }
    }
}

Basically, it does the following: Given a list of pointers to geometries, it creates two iterators within two loops. The second iterator always starts as a copy of the first; then both travel until the end of the list. If the geometries the two pointers point to touch, but do not cross (and also share some tags), they are united.

To illustrate what it does, I've drawn a very simple picture featuring five line strings which could be a possible result of my example query outlined above. In this picture, only the three grey ones are merged:

LineMerger: The grey line strings are merged, the blue ones not.

I now want to improve this approach by (a) possibly improving the algorithm and (b) putting the strain on the database server, not the slower client machine.

I've two options here, but I do not know which one is the best or which actually works:

  1. Create a SQL query which does this. However, I do not know how. (Self-join? Aggregates?) If someone could suggest a query that does this, I'd be absolutely happy.
  2. Create a PostgreSQL C extension that implements my algorithm. If so, I'm happy for any suggestions of how to improve the algorithm or improve it by enabling parallel processing.
3

This http://postgis.net/workshops/postgis-intro/advanced_geometry_construction.html and this http://blog.cleverelephant.ca/2010/07/network-walking-in-postgis.html provide the basics of recursively walking a collection.

  • My similar solution had a lot more temp tables and procedures ( but i had to do it in MS SQL ) – simplexio Oct 11 '13 at 19:10
  • Thanks! Took me some time to wrap my head around it, but that was definitively the way to go! – Technaton Nov 4 '13 at 18:28
0

See ST_LineMerge. Example:

SELECT ST_AsText(ST_LineMerge(
ST_GeomFromText('MULTILINESTRING((-29 -27,-30 -29.7,-36 -31,-45 -33),(-45 -33,-46 -32))')
        )
);
st_astext
--------------------------------------------------------------------------------------------------
LINESTRING(-29 -27,-30 -29.7,-36 -31,-45 -33,-46 -32)
(1 row)
  • Thanks; however, ST_LineMerge merges lines I already know need merging. But I've got a collection of lines which I first have to examine for possible merges. I.e., the if clause of my algorithm. Sorry if that wasn't clear; I'll edit my original post to make it clearer. – Technaton Oct 11 '13 at 17:13
0

SELECT ST_Linemerge(ST_Collect(geom)) as geom . voltage FROM table as x Group by voltage , that would try merge all lines with same voltage , or if it fails it would create or if it fails it would create one multilinestring per voltage.

Another way would be creating simple topology using PgRouting Grass example and writing something similar to this parent-child recursive query. Then use source and target column data to solve longest lines . This assumes that you want join only lines from start -> start , start -> end or end-> end . (Pgrouting may have some useful functions already done)

parent-child query is pretty simple, limit initial data to one voltage, start "walk" to tree. if call ends to situation where it would split to two paths , store existing line and start walk two new lines. I have done similar implementation to solve still connected "sinks" in network when one or more connection edges are taken out.

Performance in recursive SQL is not that good, but probably it fast enough. After you have longest lines and network only from line start -> end points, you can start to solve intersecting and touching problems ( do you need to split lines and start to create connectivity between them...)

If SQL sound too hard and you have money to spend, i recommend FME (ww.safe.com) It's nice toolbox to own when you do GIS ( file type conversion and data manipulation )

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.