A look at pedestrian infrastructure globally
Mapillary has long been a tool to derive map data from street-level images. This has spurred many creative use cases such as the mapping of canals in Amsterdam, old trails in Madeira, and cycling lanes in Medellín. The majority of imagery, however, has been road centric, as this is the most practical and scalable way to collect street-level imagery. In recent months, we’ve been exploring map data collection from a pedestrian-centric perspective, and this has highlighted new data collection opportunities, at the same time highlighting deficiencies with the OpenStreetMap dataset. This blog post is a closer look at some cities around the world to see how pedestrian data differs, and how closely it mirrors the state of pedestrian infrastructure in that city. API.
Downloading our pedestrian datasets
The starting point is to download consistent pedestrian data from OpenStreetMap across various city types. To do this, I am including a number of key=value pairs in the category of “pedestrian data”.
Using Overpass Turbo to query the OSM API, I am looking for:
- Ways that are specifically a footway
- Ways that have a footway adjacent
- Ways where foot traffic has been explicitly permitted
- Areas where pedestrian traffic is designated
- Nodes that relate to footways
This should capture most OpenStreetMap attributes that would indicate connectivity from a pedestrian-centric perspective.
You can see the exact queries included above
Did I miss any OpenStreetMap tags that are important for pedestrians? If so, let me know.
The cities we’re looking at are:
- Folsom, CA, USA
- Heidelberg, Germany
- Melbourne, Australia
- Stone Town, Tanzania
- Yesan, South Korea
Pedestrian data from Folsom, Heidelberg, and Melbourne as seen in QGIS
Pedestrian data from Stone Town and Yesan as seen in QGIS
These cities were selected because I’m relatively familiar with each of them, but most importantly, they represent different geographies, populations, income levels, and cultural realities.
For each of these cities, I want to take a look at the following two questions:
- What types of pedestrian data have been added in this city?
- How close does that pedestrian data reflect the reality on the ground?
Folsom, CA, USA
I chose this location because it’s the city I’m living in now. It’s a suburban environment with areas of incredible pedestrian infrastructure isolated from one another by car parks and busy roads.
As you can see from the QGIS visualisation, much of the pedestrian data seems to be centred around parks or commercial centers. There is very little pedestrian data in the residential areas of Folsom.
On the ground
Folsom’s population has grown by almost 66% since 2000, with much of this growth occurring in new housing estates. While the roads have been mapped fairly well, the vast majority of sidewalks in residential areas have been completely missed. The sequence below is just one of many examples of a neighborhood with sidewalks on both sides of the road, neither of which are reflected in OpenStreetMap.
Sidewalk infrastructure in Folsom is generally available in residential or shopping areas, but totally deficient in between. It’s a worthwhile exercise to map these sidewalks to paint a clearer picture of where sidewalk infrastructure falls short.
Heidelberg was the host of the 2019 State of the Map, the last one we had face to face. It’s also where the team behind the OpenRouteService (ORS) is located. I thought Heidelberg would be a good place to look at because it’s a well mapped, mid-sized city, but also very pedestrian friendly.
Heidelberg is very much the antithesis of Folsom when it comes to pedestrian data. You can see from the OpenStreetMap data that almost the entire city is well connected by sidewalks. There are also plenty of crosswalks at intersections throughout the city, but a few gaps appear, with Weststadt one such example.
Most of the sidewalks have been mapped as separate ways, although
sidewalk=both has been applied in some cases. Impressively, many internal pedestrian areas have been mapped as well in places like parks, university grounds, and hospitals.
On the ground
The OpenStreetMap community in Germany is well established and takes great pride in this shared project to map the world. Heidelberg is one of the best embodiments of this with many details across various categories being mapped.
This reflects the reality on the ground where sidewalks are not only present in most places, but well connected to one another. The main areas of improvement would be to map sidewalks as separate ways where possible and demarcate crosswalks, both of which will help pedestrian routing. One such example is on Kleinschmidtstraße, part of which has been tagged as
foot=yes and separate sidewalk ways would make pedestrian access clearer.
Melbourne has the largest population of the cities being analyzed here. It’s also my hometown, and thus the city I’ve walked the most. The area downloaded here contains the Central Business District (CBD), which is the most densely populated part of the city.
Rainy Melbourne footpath
The downtown area in focus shows a density of pedestrian relevant data, getting more sparse further away from the center. Sidewalk ways have been mapped separately, forming a navigable network connected by the crosswalks that you see in orange. Many of the crosswalks have been mapped as ways with the tag
footway:crossing rather than as individual nodes.
Funnily, the dataset I downloaded included a way from Melbourne to Tasmania across Bass Strait. This is because my download area happened to include The Spirit of Tasmania, a ferry which has been tagged as
On the ground
Overall, the central part of Melbourne is fairly well mapped, but the suburbs surrounding it become a patchwork of pedestrian data. In the CBD, there are large sidewalks and pedestrian friendly zones and these have been reflected well in OpenStreetMap. Further improvement would come by mapping crosswalk nodes and pedestrian ways with tags such as
crossing=traffic_signals to give pedestrians a clear picture of city navigability.
The areas adjacent to the CBD are well connected with sidewalks, but these are not reflected in OpenStreetMap. Adding these sidewalks and linking them with existing ones would be a good starting point for these surrounding areas.
Stone Town, Tanzania
Stone Town is the old part of Zanzibar City, the largest city on the island of Zanzibar, an autonomous region of Tanzania. The download area here focuses on this historic old town which is a fascinating maze of beautiful stone buildings. Cars are only allowed to drive a handful of roads in this area which is a welcome relief from many other cities around the world.
The view over Stone Town
One can’t look at mapping in Zanzibar without mentioning the Zanzibar Mapping Initiative (ZMI). ZMI is a collaboration between the State University of Zanzibar (SUZA), the government of Zanzibar, The World Bank and others. The initiative is notable for its use of drones and OpenStreetMap, with students of SUZA mapping the entire island with low cost drones over 2016 and 2017.
Looking specifically at Stone Town, it’s clear that the majority of pedestrian areas in the old town have been mapped. Most of the streets in Stone Town have been tagged as
highway=pedestrian, with a few of the larger, vehicle bearing ways tagged as
highway=residential. Forodhani Park near the waterfront has ways tagged as
highway=footpath, which is the only place in Stone Town you see the use of this tag. There are also areas tagged as
area:highway=pedestrian which represent some of the pedestrian friendly town squares among the densely packed buildings.
One thing you might notice looking at the data is that Stone Town’s pedestrian network exists as an island from other parts of the city. We’ll take a look at whether this matches reality below.
On the ground
If we are to focus only on the downtown area mentioned above, then Stone Town is fairly well mapped from a pedestrian’s perspective. ORS had no trouble navigating some of the start/end combinations I threw at it as most of the pedestrian navigable paths have been added to the map.
As mentioned above however, this area exists as an island separated from other parts of the city. This is only partially true in practice. The are definitely pedestrian linkages to the rest of the city, but these are not fully reflected in OpenStreetMap. Creek Road which is a major thoroughfare running northeast to southwest seems to divide the old town from Zanzibar City as a whole.
The 360º sequence below is just one example of a sidewalk linking the two parts of the city which is not reflected on the map. The well maintained sidewalk here contrasts with the section further south where Creek Road intersects with Market Street. This area is heavily trafficked by visitors to both the market and the bus station. As such, both the sidewalk and road are in need of maintenance.
Yesan, South Korea
Yesan is a relatively small city a little over 2 hours south of Seoul. It serves as the county headquarters for Yesan County and a hub for a lot of the agricultural activity in the area. Unlike some of the other places on this list, mapping in South Korea is dominated by local players such as Naver. This is mainly attributable to government regulation and innovative local companies, so it makes it all the more interesting to take a look at OpenStreetMap data in Yesan.
Pedestrian data in Yesan is almost non-existent. One trail had been mapped by the river, but there was very little else relevant to pedestrians. I added a few trails and crosswalks using satellite imagery and my recollection of the town, but there is still an enormous amount of work to do.
One of the fun things about mapping in Korea though is that there is still so much to be mapped. Even major roads outside Yesan’s County offices were missing, a reality that is probably emblematic of OpenStreetMap across much of Korea. It’s to be expected that pedestrian relevant data is also missing for most of the country.
On the ground
One of the remaining agricultural areas in Yesan
Yesan is a very pedestrian friendly city, with sidewalks adjacent to most roads and pedestrian crossings throughout. There are also a number of walking trails traversing the city’s parks. The exception to this are the smaller, narrower roads that are usually found in the older part of town. Fortunately the speed-limits are lower on these roads, so improving pedestrian data would extend to mapping these smaller residential roads and their speed-limits. It’s also worth adding the tag ‘foot=yes’ to these roads to make it clear both vehicles and pedestrians are welcome. Lastly, there are a number of walkways that run behind houses, connecting them with a nearby road. These walkways are very difficult to see from satellite imagery, but could easily be mapped with a 360º camera or smartphone.
Reflecting the reality on the ground
It’s quite fascinating looking at how different cities prioritize various modes of transport. Cities such as Heidelberg are incredibly pedestrian friendly, while others such as Folsom vary widely as you move through the city. This exploration of cities around the world drew our attention to quite a few different types of pedestrian data. These differences can be explained by a variation in tagging conventions, the completeness of data, and the types of pedestrian infrastructure in each city. Reducing the gaps for the first two factors can bring us closer to understanding the reality on the ground and future conversations about pedestrian infrastructure.