Skip to content

Performance

Can Yang edited this page May 30, 2019 · 1 revision

Map matching

A case study is reported in the original paper with real world datasets in Stockholm:

  • GPS: 644,695 trajectories containing 6,812,720 points
  • Road network: 23,921 nodes and 57,928 directed edges
  • UBODT size: 4,305,012 rows
  • k = 8 (candidate set size), r = 300 meters (search radius)

As reported in the paper, on a desktop with Intel(R) Core(TM) 2 Quad CPU Q9650 @ 3.00GHz (4 processors), the speed of map matching (single processor) is about:

  • 25000 points/s (WKB Geometry output, mode 1)
  • 45000 points/s (No geometry output, mode 0)

On Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz (8 processors), the speed are reported as:

Program WKB speed WKT speed No geom speed
fmm 52,890 17,905 58,830
fmm_omp 221,797 77,910 289,031

UBODT precomputation

Statistics of ubodt construction on the network of Netherland (700k nodes and 1 million edges) http://geodata.nationaalgeoregister.nl/nwbwegen/extract/nwbwegen.zip

delta (m) running time rows csv size
1000 3min27.5s 19,022,620 766 MB
3000 4min32.9s 79,998,367 3,2GB