Monday, 4 May 2015

Documenting Footpaths with Mapillary

I have long been a believer in the need to document OpenStreetMap survey data as thoroughly as possible.

I have a large archive of audio files, GPS traces, and tens of thousands of photographs. These span back to late 2008 when I started contributing to OSM. From time to time these prove useful, for instance, I had very precise documentation for my evidence at a Public Enquiry.

However, sharing such archival information with other mappers is difficult. It's not even straightforward for me to locate stuff. I have used OpenStreetView (OSV) since it was announced at SotM Girona. It is difficult to share photos using OSV, and the interface has not developed since 2010.



I was therefore very interested to learn about Mapillary, but was initially put off by the licensing. When they changed the licensing I was more interested. At SotM-EU Karlsruhe I was able to chat with Yubin after hearing his talk, which convinced me to give it a go. As I've said before, I regret I did not do this the following morning when full documentation of our walk at the Weingartnermoor would have been very useful, not just for mapping this particular place, but for discussing how to map woodland.

I don't have an Android phone which is compatible with Mapillary so I have had to do things manually. This is a little tedious, so I tend to keep the creation of sequences for things which are either simple or of particular value.


I only really started creating decent Mapillary sequences just before going to Buenos Aires for State of the Map. Peter Neubauer showed some great examples from the cemetery at Recoleta.

I made quite a few sequences in Ushuaia, El Calafate, Puerto Madryn, Bariloche and Buenos Aires. Sadly bandwidth was rarely enough to successfully upload them whilst I was there. One of the things I liked about using Mapillary in Argentina is that it provides a different way to share what a place is really like.

I didn't take photos of too many places outside of cities because I was usually carrying a fair bit of other equipment, but I did make two useful sequences: on the viewing walkways of the Perito Moreno Glacier (above), and in some Guindo (Nothofagus pumilio) woods at, outside Ushuaia.

On my return to the UK I decided I would like to try and capture longer sequences on local footpaths.

Creating sequences manually involves a fair bit of manipulation of the photos:
  • Geotagging. I have used Geosetter for several years, and normally geotag photos immediately after download.
  • Adding View Directions. Geosetter allows me to do this, but this was new to my workflow. Doing it for several hundred photos is tedious, so I normally just add a rough sense of direction to large groups of photos. This tends to mean I avoid taking panoramas.
  • Resizing Photos. I use ImageMagick to reduce the size of images to 2400x1800 and lower the quality a bit. More recently I also auto-adjust the gamma to handle overly bright or dark pictures. Reducing photo size really improves how rapidly they get processed by Mapillary. I think the settings I use are a decent compromise and retain enough of the key detail.
  • Manual Upload. The upload process has its faults (and has recently been refactored), but I generally just persist. Smaller file size helps a lot.
One thing I think OSV did well was the user moderation, and I think Mapillary will have to move in this direction as the volume of uploads gets too large for the development team. Slower moderation means more work needs to be done on images before upload: for instance altering view direction is too tedious to do on any scale online.
I've used Mapillary for a number of purposes:
But, most of all I have used it for recording footpaths.

Mapping footpaths in Britain is a big task. We now expect to collect a lot of detailed information on paths, and the only practical way to collect this detail is to walk the paths. In practice some of the detail gets omitted when a path is first added to OSM: one wants to get the main information in first.

Different people have various ways of collecting footpath data. Nick Whitlegg uses his own OpenTrail app on an Android Phone; SomeoneElse has an elaborate method for collecting footpath data refined over many years: but I have always tended to rely on geotagged photos. Mapillary really provides a more structured way to collect and store such photos.

The downside is that I am using a camera rather than a smart phone for collecting images. A bit of experimentation shows that taking pictures every 10 or 20 paces works well: one can vary the frequency depending on the scenery. Carrying the camera makes surprisingly little impact on the speed I walk, my ability to notice other things, or carry on conversations. In fact the most tedious thing is that I use different settings on the camera for Mapillary sequences compared with wildlife photos.

Processing photos involves geotagging them from my GPS traces using Geosetter: something I have done for years. Additionally Mapillary needs direction information: adding this takes more time, although I often just assign a general direction to a whole group of photos. In principle this data can be changed once the sequence is stored by Mapillary, but the workflow for this is still too clunky. It is best to spend the time getting it reasonably accurate up front.

I also pre-process the images further: I reduce their size, quality and sometimes auto-adjust the gamma using ImageMagick. This tends to lead to faster processing of images with fewer hiccoughs. I imagine some of these steps, particularly auto adjusting brightness and contrast might become part of the server-side processing by Mapillary. An image size of 2400x1800 keeps nearly all the useful information from larger images.

I find I prefer to use Mapillary to find locations in my own photos for editing over the tools I have on my desktop. Of course, even better, others can use this information to add things like footpath fingerposts, numbers of steps on stiles, or check views of the countryside before armchair mapping from aerial imagery.

So far I have made 5-6 sets of sequences of footpaths:



  • Petworth area West Sussex. This was my first serious experiment. I joined Nick Whitlegg again to continue mapping paths in the Weald. I thought I might make some fairly short sequences, but in practice I continued throughout the day (about 5.5 hours of walking). A particularly nice feature of this walk is that we crossed the path of one we did last year. The sequence therefore includes a couple of vineyards!
  • Gringley-on-the-Hill, Nottinghamshire. The following weeked I was joined by SomeoneElse & Mot in North Nottinghamshire, our first attempt to do something similar to what Nick has been doing in Southern England. Gringley is a very attractive village full of 18th-19th century brick houses with pan-tiled roofs. As the name implies it sits on a low, but prominent ridge. To the North fen country starts, extending towards The Humber. These flat areas were drained by Dutch engineeers around 250 years ago, and are known as the Carrs (from an English word associated with wet woodland of Alder and Willow). South of the ridge is more rolling country with the Chesterfield Canal a prominent feature. We did two walks: one in the morning S towards the canal, with each person returning by a different route. In the afternoon we walked companionably together to the South East. Separately, SomeoneElse and I also did a little bit of waling in the Carrs before we met. Following on from Rob Nickersons demonstration of using umap for our Fauld Crater meeting, I have created a similar map showing what we did.
    See full screen

  • Leicestershire Village paths. Many villages have quite a network of short paths which it is difficult to map as part whilst mapping paths between villages. I picked up a few in Gringley at the end of our meeting. A week or so later I made a short trip into Leicestershire and spent time surveying three villages: Hoton, Wymeswold, and Rempstone. Adding things like addresses would have taken more time, but I was able to sort out quite a number of paths (although many in Wymeswold, although marked on the rowmaps data, are not signposted as public rights of way on the ground). This type of Mapillary sequence is very useful for detailed mapping of villages as well.
  • Plaistow area, West Sussex. Another Wealden walk with Nick. This was an attempt to penetrate an area of the county which is very poorly served by public transport. I only documented thte morning session because I discovered I cant delete photos from the camera once I use more than 4G and I was not carrying a spare micro-SD card. The countryside is a little less interesting than the Greensand ridge to the South (and no vineyards), but this was compensated by a range of interesting flowers (Early Purple Orchids, Bluebells, Wood Sorrel, Wood Spurge, Goldilocks Buttercup, Dog Violets) and a tree I very rarely see, the Wild Service Tree.


  • Bisham Woods. The aim of this walk was to document both footpaths and woodland. The walk largely covered the ground I discussed in my post on Spring Woodland last year. The time was about 6 weeks later, so the woodland canopy had closed, and the bluebells were out. So this is one way to share a typical British woodland scene of this time of year. I must repeat the walk next year before the canopy closes: one thing that Mapillary does enable is looking at phenology (Nature's Calendar).

Conclusions

Mapillary enables lots of details of footpaths and the surrounding countryside to be recorded. Even using a hand-held camera and manual editing of the EXIF data it is not too arduous to upload a day long walk. In fact this may avoid some of the issues Malenki experienced with his very ambitious recording of 3 weeks walking in Albania.

There exists considerable potential for sharing information about the state of footpaths, and not just for mapping. I have tried to use some of the images to report problems with specific paths to local councils already. Similarly, leaders of walks could benefit from being able to get an idea of a given path before walking it.

In general, when I remember, I have tried to take good photographs of barriers. The motivation comes very much from a query on OSM Help. However well we document stiles and other barriers to walkers ultimately some people still need a photograph to be able to judge whether it is appropriate for them. This process can be helped by using a mapillary tag directly on OSM (although in the main I use this as a kind of source tag). So this isn't directly an answer to the query, but it is moving the idea forward.

I am using Mapillary sequences for far more than just mapping. I find them more personal than something like Google StreetView. Obviously they have other advantages in that its possible to create them whereever a small camera can go, they can be created quickly, multiple time views are not just possible, but desirable. There is also massive scope for experimentation : one thought I have is very detailed sequences for good gardens. Equally the strengths of the Mapillary team in image processing and recognition gives promise of more things to come.

2 comments:

  1. Hi Jerry,

    I've seen many photos made by you when I was playing the mapillary traffic game! I like to trace footpath as well. Brasília has many footpaths and despite the fact that the city is huge and it has long distances between the sectors, it's a very nice city to walk. So I try to map the footpaths and make the foot routes calculation better.

    ReplyDelete
  2. Funnily enough just as I was finishing off this post someone from Brazil posted a diary entry on using Mapillary in the countryside too: http://www.openstreetmap.org/user/Edil%20Queiroz%20de%20Araujo/diary/34918. Glad to hear Brasilia is a walker friendly city, I suppose that is one benefit of the way it was laid out.

    ReplyDelete