Tools for point-of-interest (POI) extraction, walkability/attractiveness indexes and tiling of XML map data
OSMToolset package provides the tools for efficient extraction of point-of-interest from maps and building various custom walkability indexes in Julia.
Documentation: 
using Pkg; Pkg.add("OSMToolset")- Export points-of-interests (POIs) from a OSM xml map file to a
DataFrame - A spatial attractiveness index for analyzig location attractivenss across maps (can be used for an example in research of city's walkability index)
- A spatial index for finding nearest nodes in maps to a given
LLAorENUcoordinates - OSM map tiling/slicing - functionality to tile a large OSM file into smaller tiles without loosing connections on the tile edge. The map tiling works directly on XML files
(a complete code for this visualization can be found in the docs)
Please note that the maps provided by the OpenStreetMap project contain very detailed information about schools, businesses, shops, restaurants, cafes, parking spaces, hospitals etc. With this tool you get an effient, customizable API for extraction of data on such points of interests for further processing. This information can be further used e.g. to build walkability indexes that can be used to explain attractiveness of some parts of a city. Hence the second functionality of the package is to provide an interface (based on the SpatialIndexing package) for building of efficient attractiveness indexes of any urban area.
Since the OSM map XML files are usully very large, sometimes it is required to tile the files into smailler chunks for efficient parallel processing. Hence, yet another functionality of this package is an OSM file tiler.
This toolset has been constructed with performance in mind for large scale scraping of spatial data. Hence, this package should work sufficiently well with datasets of size of entire states or countries.
The examples assume that the sample file is used
file = sample_osm_file()
Let us use the default configuration for parsing.
julia> df1 = find_poi(file)
78ร10 DataFrame
Row โ elemtype elemid nodeid lat lon key value โฏ
โ Symbol Int64 Int64 Float64 Float64 String String โฏ
โโโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
1 โ node 69487440 69487440 42.3649 -71.1029 public_transport stop_positi โฏ
โฎ โ โฎ โฎ โฎ โฎ โฎ โฎ โฎ โฑ
78 โ relation 7943642 2913461577 42.3624 -71.0847 leisure park โฏ
4 columns and 76 rows omitted
The default configuration file can be founds in OSMToolset.__builtin_config_path. This configuration has meta-data columns that can be seen in results of the parsing process. You could create on base on that your own configuration and use it from scratch.
Suppose that rather you want to configure manually what is scraped. Perhaps we just wanted parking spaces
that can be either defined in an OSM file as amenity=parking or as parking key value:
julia> config = ScrapePOIConfig("parking",("amenity","parking"))
ScrapePOIConfig{NoneMetaPOI} with 2 keys:
No โ key values
โโโโโผโโโโโโโโโโโโโโโโโโ
1 โ amenity parking
2 โ parking *
Note that the scraping configuration can be extracted to a data frame by executing config |> DataFrame. Such dataframe can also be used to create a new configuration by executing ScrapePOIConfig{NoneMetaPOI}(DataFrame(key=["amenity","parking"],values=["parking","*"])).
Note that since we do not use meta data yet we use parameter: NoneMetaPOI.
Now this can be scraped as :
julia> df2 = find_poi(file, config)
12ร7 DataFrame
Row โ elemtype elemid nodeid lat lon key value
โ Symbol Int64 Int64 Float64 Float64 String String
โโโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
1 โ way 187565434 1982207088 42.3603 -71.0866 amenity parking
โฎ โ โฎ โฎ โฎ โฎ โฎ โฎ โฎ
12 โ way 1052438049 9672086211 42.3624 -71.0878 parking surface 10 rows omitted
It is also possible to extract adjacent tags within the same node - this cab be achieved via the all_tags option.
For an example we could get the information on parking place metadata.
find_poi(file, ScrapePOIConfig("parking",("amenity","parking")); all_tags=true)
25ร7 DataFrame
Row โ elemtype elemid nodeid lat lon key value
โ Symbol Int64 Int64 Float64 Float64 String String
โโโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
1 โ way 187565434 1982207088 42.3603 -71.0866 amenity parking
2 โ way 187565434 1982207088 42.3603 -71.0866 access private
3 โ way 187565434 1982207088 42.3603 -71.0866 parking surface
4 โ way 187565434 1982207088 42.3603 -71.0866 surface asphalt
โฎ โ โฎ โฎ โฎ โฎ โฎ โฎ โฎ
25 โ way 1052438049 9672086211 42.3624 -71.0878 parking surface
20 rows omitted
It can be seen that the same nodeid is repeated for different tags.
The data that we extract can be decorated with additionaly information, such as range and influence of the POI.
julia> config2 = ScrapePOIConfig(("amenity","cafe")=>AttractivenessMetaPOI(:food,1,500), ("amenity","restaurant")=>AttractivenessMetaPOI(:food,2,1000), ("parking",("amenity","parking")) => AttractivenessMetaPOI(:car,1,500))
ScrapePOIConfig{AttractivenessMetaPOI} with 2 keys:
No โ key values group influence range
โโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
1 โ amenity cafe food 1.0 500.0
2 โ amenity restaurant food 2.0 1000.0
Here we assume that the importance of restaurant is larger than of cafe and that people are more likely to walk a larger distance to visit a restaurant.
julia> filter!(r->r.nodeid in [1884055322, 11173231405], # select two places
find_poi(file, config2, all_tags=true))
5ร10 DataFrame
Row โ elemtype elemid nodeid lat lon key value group influence range
โ Symbol Int64 Int64 Float64 Float64 String String Symbol? Float64? Float64?
โโโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
1 โ node 1884055322 1884055322 42.3617 -71.09 amenity cafe food 1.0 500.0
2 โ node 1884055322 1884055322 42.3617 -71.09 name Forbes Family Cafe missing missing missing
3 โ node 1884055322 1884055322 42.3617 -71.09 opening_hours Mo-Fr 11:00-15:00 missing missing missing
4 โ node 11173231405 11173231405 42.3622 -71.0864 amenity cafe food 1.0 500.0
5 โ node 11173231405 11173231405 42.3622 -71.0864 name Ripple Cafe missing missing missing
The data can be further processed in many ways. For example here is a sample code that performs POI vizualisation
Let's consider a more complex attractiveness information:
config3 = ScrapePOIConfig(("amenity","cafe")=>AttractivenessMetaPOI(:food,1,500), ("amenity","restaurant")=>AttractivenessMetaPOI(:food,2,1000), (["parking",("amenity","parking")] .=> Ref(AttractivenessMetaPOI(:car,1,500)))... )
ScrapePOIConfig{AttractivenessMetaPOI} with 4 keys:
No โ key values group influence range
โโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
1 โ amenity cafe food 1.0 500.0
2 โ amenity parking car 1.0 500.0
3 โ amenity restaurant food 2.0 1000.0
4 โ parking * car 1.0 500.0
Note that in this demo we assume attractiveness configuration defined as AttractivenessMetaPOI. If you want a different structure of data for this index you need to crate a subtype of MetaPOI and use it in the constructor.
We search for such locations:
julia> df3 = find_poi(file, config3)
18ร10 DataFrame
Row โ elemtype elemid nodeid lat lon key value group influence range
โ Symbol Int64 Int64 Float64 Float64 String String Symbol Float64 Float64
โโโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
1 โ node 1884054889 1884054889 42.3621 -71.0892 amenity cafe food 1.0 500.0
2 โ node 1884055322 1884055322 42.3617 -71.09 amenity cafe food 1.0 500.0
โฎ โ โฎ โฎ โฎ โฎ โฎ โฎ โฎ โฎ โฎ โฎ
17 โ way 1052438049 9672086211 42.3624 -71.0878 amenity parking car 1.0 500.0
18 โ way 1052438049 9672086211 42.3624 -71.0878 parking surface car 1.0 500.0
14 rows omitted
Now with this data we create a spatial attractiveness index in the following way:
ix = AttractivenessSpatIndex(df3);
Let us consider a point on the map:
using Statistics
lat, lon = mean(df3.lat), mean(df3.lon)
We can use the API to calculate attractiveness of that location:
julia> attractiveness(ix, lat, lon)
(car = 8.595822085195946, food = 5.151440338789913)
For this location we can see it is easy to find food and park your car nearby.
If, for some debugging purposes, we want to understand what data has been used to calculate that attractiveness use the explain=true parameter:
julia> attractiveness(ix, lat, lon; explain=true)
(car = 8.595822085195946, food = 5.151440338789913, explanation = 18ร7 DataFrame
Row โ group influence range attractiveness poidistance lat lon
โ Symbol Float64 Float64 Float64 Float64 Float64 Float64
โโโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
1 โ food 1.0 500.0 0.183414 408.293 42.3599 -71.0913
โฎ โ โฎ โฎ โฎ โฎ โฎ โฎ โฎ
18 โ food 2.0 1000.0 1.44716 276.42 42.3627 -71.084
16 rows omitted)ted
The attractiveness function is fully configurable on how the attractiveness is actually calculated. The available parameters can be used to define attractiveness dimension, aggreagation function, attractivess function and how the distance is on map is calculated.
Let us for an example take maximum influence values rather than summing them:
julia> att = attractiveness(ix, lat, lon, aggregator = x -> length(x)==0 ? 0 : maximum(x))
(car = 0.8840868352005442, food = 1.747669233262405)
We could also used a DataFrame without meta data columns for the attractiveness:
df4 = find_poi(file, ScrapePOIConfig(("amenity","parking"), "parking"))
ix4 = AttractivenessSpatIndex{NoneMetaPOI}(df4; get_range=a->300, get_group=a->:parking);
Note that since we did not have metadata we have manually provided 300 meters for the range and :parking for the group.
Now we can use this custom scraper to query the attractiveness:
julia> attractiveness(ix4, lat, lon; aggregator = sum, calculate_attractiveness = (a,dist) -> dist > 300 ? 0 : 300/dist )
(parking = 30.235559263812686,)
Note that for this code to work we needed to provide the way the attractiveness is calculated with the respect of metadata a (now an empty struct as this is NoneMetaPOI).
The native format for OSM files is XML. The files are often huge and for many processing scenarios it might make sense to slice them into smaller portions. That is where this functionality becomes handy.
The file tiling can be executed as follows:
outfiles = tile_osm_file("file.osm", nrow=2, ncol=3, out_dir="some/target/directory")
After the execution outfile will be a matrix with file names of all tiles.
The OSM tiler is simultanously opening a file writer for each file. The operating system might limit the number of simultanously opened file descriptors. If you want to create large number of tiles you need to either change the operating system setting accordingly or use a recursive approach to file tiling.
This research was funded by National Science Centre, Poland, grant number 2021/41/B/HS4/03349.
This tool is using some code from the previous work of Marcin ลปurek, under the same research grant. The initial prototype can be found at: https://github.com/mkloe/OSMgetPOI.jl