Cross Street Indexer

Blazing fast tile based geocoder that matches cross street (road intersections) entirely sourced by OSM QA Tiles.

Features

• Blazing fast 1/20th of a millisecond search (275,000 ops/sec)
• Processed United States OSM QA Tiles in 24m 27s (189714 tiles)
• Reads streaming index data
• Easy to use CLI to create & search index
• NodeJS 6 & 7 compatible
• Only uses Tile Reduce + Turf
• Indexes can be published on S3 buckets
• Bundled 5MB QA-Tiles for testing purposes

Process

• Step 1: Filter data from QA-Tiles (lib/qa-tiles-filter.js)
• Step 2: Extract road intersections from QA-Tiles (lib/intersections.js)
• Step 3: Normalize street name ABBOT AVE. => abbot avenue (lib/normalize.js)
• Step 4: Convert intersections into multiple points using a combination of road & ref tags (lib/geocoding-pairs).
• Step 5: Group all hashes into single Quadkey JSON object (lib/reducer.js)
• Step 6: Generate index cache from QA Tiles via CLI (bin/cross-street-indexer.js)
• Step 7: Stream or read from disk index caches via CLI (bin/cross-street-search.js)
• Step 8: (optionally) Publish to S3 to share the index

OSM QA Tiles

Users can download the entire planet or country extracts of OSM QA Tiles from https://osmlab.github.io/osm-qa-tiles.

Install

npm

$ npm install --global cross-street-indexer

yarn

$ yarn global add cross-street-indexer

Quickstart

$ cross-street-indexer united_states_of_america.mbtiles
$ cross-street-search "Chester St" "ABBOT AVE."
-122.457711,37.688544

CLI

Cross Street Indexer

$ cross-street-indexer --help

  Cross Street Indexer

  Usage:
    $ cross-street-indexer <qa-tiles>
  Options:
    --output    [cross-street-index] Filepath to store outputs
    --bbox      Excludes QATiles by BBox
    --tiles     Excludes QATiles by an Array of Tiles
    --debug     [false] Enables DEBUG mode
  Examples:
    $ cross-street-indexer latest.planet.mbtiles
    $ cross-street-indexer latest.planet.mbtiles --tiles [[654,1584,12]]
    $ cross-street-indexer latest.planet.mbtiles --bbox [-122.519,37.629,-122.168,37.917]

Cross Street Search

$ cross-street-search --help

  Cross Street Indexer

  Usage:
    $ cross-street-search <name1> <name2>
  Options:
    --output    [cross-street-index] filepath to Cross Street index output folder
    --tiles     Lookup index files via an Array of Tiles or Quadkeys
    --bbox      Lookup index files via BBox
    --latlng    Outputs LatLng instead of the default LngLat
    --stream    Enables reading from streaming index file (ignores tiles/bbox options)
  Examples:
    $ cross-street-search "Chester St" "ABBOT AVE."
    $ cross-street-search "Chester" "ABBOT"
    $ cross-street-search "Chester St" "ABBOT AVE." --tiles [[654,1584,12],[653,1585,12]]
    $ cross-street-search "Chester St" "ABBOT AVE." --tiles "023010221110,023010221110"
    $ cross-street-search "Chester St" "ABBOT AVE." --bbox [-122.5,37.6,-122.1,37.9]
    $ cat 023010221110.json | cross-street-search "Chester St" "ABBOT AVE."
    $ curl -s https://s3.amazonaws.com/<your-s3-bucket>/latest/023010221110.json | cross-street-search "Chester St" "ABBOT AVE." --stream

Normalization Process

Normalization should follow the following standards:

• Drop any period if exists
  • ave. => avenue
• Street suffix to full name
  • ave => avenue
  • CIR => circle
  • ln => lane
  • HWY => highway
• Name should be entirely lowercase
  • Parkside Avenue => parkside avenue
• Direction to full word
  • N => north
  • S => south
  • NE => northeast
• Numbered street names to Abbreviations
  • first => 1st
  • third => 3rd
  • ninth => 9th
  • fifth => 5th
• Remove any additional information
  • rodeo avenue trail (dead end ford bikes--no bikes on 101) => rodeo avenue trail

Index -- JSON Lines

The Cross Street Index is stored in an easy to read key/value JSON Lines format.

• key: Normalized road pairs (<name1>+<name2>)
• value: Longitude & Latitude (Array<number>)

{"abbot avenue+chester street":[-122.457711,37.688544]}
{"chester street+abbot avenue":[-122.457711,37.688544]}
{"chester street+lisbon street":[-122.45821,37.68796]}
{"lisbon street+chester street":[-122.45821,37.68796]}
{"hoffman street+lisbon street":[-122.456764,37.687179]}

Index -- LevelDB

The LevelDB storage is very similar to the JSON Line format with & without the added Quadkey to the key.

• key: Normalized road pair (<quadkey>+<name1>+<name2> or <name1>+<name2>)
• value: Longitude & Latitude (String -- <lng>,<lat>)

"023010221110+abbot avenue+chester street" => "-122.457711,37.688544"
"023010221110+chester street+abbot avenue" => "-122.457711,37.688544"
"023010221110+chester street+lisbon street" => "-122.45821,37.68796"
"023010221110+lisbon street+chester street" => "-122.45821,37.68796"
"023010221110+hoffman street+lisbon street" => "-122.456764,37.687179"

Design Considerations

• NodeJS Support 4 & 5 6 & 7
• Multi-lingual tagging name:en & name:fr, etc...
• Secondary Link breaking road intersections
• Match road intersections with/without road suffixes:
  • abbot avenue+chester street
  • abbot+chester
  • abbot avenue+chester
  • abbot+chester street
• Normalized Highways use short names:
  • HWY 417 => highway 417
  • Highway 417 => highway 417
  • ST 130 => st 130
  • CA 130 => ca 130
• ?? Consider using ! instead of + as separator (less common to have a ! in a name)
• ?? Consider using abbreviation suffixes:
  • Index cache file size reduced
  • Easier to normalize road names
  • Lowercross-street hashes combinations
  • Easier to match major highways
• Add extra name to numbered highways ref & name ("CA 131" => ["ca 131", 131])
• Add extra hash combination with the removal of street suffix (rodeo avenue => [rodeo avenue, rodeo])
• Does not save empty z12 cross street indexes (reduces total number of files).
• Extra \n at the bottom of the file (helps concatenate streams together).
• Split name & ref code with ; into Array of names.
• Search output has \n at the end (-122,37\n or -122,37)
• Loops create conflicts if cross street is in the same z12 tile.
• Turning Circles without any names are exclude.
• Loading more than 1 index cache might result in loss of data, however these conflicts are very minimal (ex: only 0.2% conflicts using 4 San Francisco tiles)

OSM Attributes

• name: Street name (Abbot Avenue)
• ref Reference number normaly used to tag highways numbers
• highway classification (residential, primary, secondary)
• bridge yes/no
• tunnel yes/no
• @id OSM ID of street (way)

Debugging

Including --debug will store additional files for each QA-Tile which can be helpful for debugging.

• <output>/<levedb> - LevelDB of all cross streets combined
• <output>/<quadkey>.json - Cross Street index cache
• <output>/<quadkey>/highways.geojson - Highways only
• <output>/<quadkey>/intersects.geojson - Point which intersect roads
• <output>/<quadkey>/names.json - All unique names & ref
• <output>/<quadkey>/stats.json - Debugging stats

debug.json

{
	"tile": [
		654,
		1584,
		12
	],
	"quadkey": "023010221110",
	"features": 49003,
	"lines": 2427,
	"intersects": 1921,
	"index": 3882
}

References

• Natural Node to create fuzzy mathes
• Carmen to normalize roads
• LevelDB as storage