Skip to content

Latest commit

 

History

History
47 lines (33 loc) · 2.28 KB

low-level-api.md

File metadata and controls

47 lines (33 loc) · 2.28 KB

Low level API

If you just start to use GraphHopper please refer to routing docs or the quickstart for developers and come back here later if the higher level API does not suit your needs.

What are pillar and tower nodes?

From road network sources like OpenStreetMap we fetch all nodes and create the routing graph but only a sub-set of them are actual junctions, which are the ones we are interested in while routing.

Those junction nodes (and end-standing nodes of dead alleys) we call tower nodes which also have a graphhopper node ID associated, going from 0 to graph.getNodes(). The helper nodes between the junctions we call 'pillar nodes' which can be fetched via edgeIteratorState.fetchWayGeometry(FetchMode.PILLAR_ONLY). Avoiding the traversal of pillar nodes makes routing a lot faster (~8 times).

That splitting into pillar and tower nodes is also the reason why there can't be a unique mapping from one OSM node ID to exactly one GraphHopper node ID. And as one OSM Way is often split into multiple edges the same applies for edge IDs too.

What are virtual edges and nodes?

For a route you do not only need junction-precision, i.e. from tower node to tower node, but we want GPS-precise routes, otherwise you'll get lots of trouble for oneways and similar.

To make GPS precise routes possible, although we route from tower node to tower node, we introduce one new virtual node x and virtual edges A-x, x-B for every query point located on an edge A-B:

\                /
 A---x---------B
/                \

But we need to decouple requests from each other and therefor we create a very lightweight graph called QueryGraph for every request which handles also stuff like two query points on the same edge.

The virtual nodes and edges have a higher int ID than graph.getNodes() or allEdges.length()

A call QueryGraph.create(graph, allQRs) will determine the correct node for all Snaps: and either create new virtual nodes or if close enough use the existing junction node.

Handle graph and routing algorithms

See this code for more details about using the low level API.