When many people point me towards something, then I know I should pretty much blog it, in an ideal world before the US wakes up too!
I’m going to start with the sound bite that normally wraps up blog posts.
This geocoder is built by the people, whether they are aware of it or not. People enter tags for photos into flickr and people enter the location of those photos via a map (or GPS data). Which is effectively mapping tags to location and in this instance the photos are incidental (that’s left to Fotoland.us). There’s no master geocoding database where the data is entered by agencies, it’s Folksonomy made useful, democratic mapping. That makes it slightly open to all the abuse that comes with that, but the shear weight of numbers should smooth things out.
Here’s a visual example from Flickr, photos tagged with Manchester.
It’s a bit blurry, but see if you can guess where Manchester is? Yup, it’s the place with a pink dot that says “Lots” over it.
The method is simple, when you ask the geocoder for a location (or object), it asks flickr for geotagged photos tagged with whatever you’re searching for, it then uses the geo-information for those photos to figure out where the location or object you’re asking for is.
How do I know this works? Because it’s being done by the Flickr Geocoder created by Tim Waters over at thinkwhere. Here’s a quick image showing what’s going on when you ask it where Manchester is.
The clustering is done following the Java k-means clustering algorithm, the final location being the center-ish-thing of the largest cluster.
As Mikel points out in this post; flickr Geocodr, this can be used for all sorts of things, not just places…
It works remarkably well .. for cities, neighborhoods (â€haight ashburyâ€, “hells kitchenâ€, “calle ochoâ€), places and landmarks (â€empire state buildingâ€, “red squareâ€). It doesn’t always work precisely (â€statue of libertyâ€), but suspect this could be down to some tweak in the algorithm.
Personally I quite like The M25 around London while Mikel suggests San Francisco Bay to Breakers as an example of it being used for detection of linear objects.
As GPS chips get embedded into cameras (and voluntarily get used), more and more, and people continue to tag, more and more, systems like this will get smarter and smarter.
So next time you take a photo, drop it on the map and tag it “rockpool”, be happy in the knowledge that you’re adding to a giant repository that’ll tell people where the coastline is.
[...] Which also ties into the Flickr Geocoder post of a couple of days ago, borders and location can now be defined by people rather than a single “in power” appointed agency. [...]
[...] Ok, so (very) basically it’s pulling the tags from geotagged photos and figuring out which tags represent what area, so in some respects it’s similar to the Flickr Geocodr I posted about a while back. Looking at the TagMap for London you can clearly see both areas and landmarks being well defined on the map. Dragging the map around and zooming in and out reveal what’s been tagged at different levels. [...]