GeoSpy + OpenStreetMap search = super power
Some of the most important details to look at first when detecting the geolocation of a photo are the street signs. Unfortunately, it is not uncommon for only a piece of one or more street signs to be visible in the photo. For example:
Here we only see that there is a certain sign that has a word written on it that ends in “bka”. And another sign that has the word “Bella” (or maybe “Billa” or “Bolla”) on it.
It’s not an easy task. But we can try to solve it using AI and Bellingcat OpenStreet Map Search.
The first thing we need to do is to define an approximate search area (ideally a city, but a country will also good). Since OpenStreetMap Search cannot search the whole world at once, but only individual areas, no larger than a single region or a small country.
How to detect the search area?
Let’s try to load this image into GeoSpy (https://geospy.web.app).
AI service suggested that the place is in Warsaw and even offered approximate coordinates.
Let’s try to search these geographic coordinates in Google Maps and see what Google Street View has to show.
Unfortunately, the service didn’t work accurately. I looked round the buildings nearest to this point and found nothing similar.
In this case, we have at least three ways of work:
- analyse different details of the image (e.g. look closely at a parking lot) and think about what other clues there might be;
- scrolling through Google Street View in the hope that the coordinates obtained do not have too much error (sometimes it really helps!);
- try searching for establishments with the right letters in the name using OpenStreetMap Search (Overpass Turbo) in Warsaw.
Unfortunately, the GeoSpy service is sometimes wrong with the city (but rarely wrong with the country) and I can’t claim that this place is exactly in Warsaw. I’m just testing a hypothesis. Hypothesis testing is a very effective technique in photo geolocation (although labour intensive).
OpenStreetMap search
Go to https://osm-search.bellingcat.com/ and set up custom feature.
OSM Key = name ends with “bka”
And let’s search in Warshaw.
We see that there are about 500 places in Warsaw called Zabka. Let’s google what the signboard of this place looks like.
Yes! It’s very similar! That’s what we need!
Very good news: we recognised the full name of the place in the picture!
Good news: the photo was definitely taken in Poland and could be taken in Warsaw.
Bad news: there are 9,000+ Zabka stores in Poland and 500 of them in Warsaw.
Theoretically, we could just look at 500 points found on Google Street View. If we looked at 10 stores per minute, we could do it in less than an hour. But we’ll keep thinking.
Let’s look at the original photo again.
We see that there is a restaurant very close to the Zabka store. We can assume that the name of the place has the word “Bella” in it’s name. But we don’t know for sure, it could be just a listing of dishes or name of singer. Let’s test the hypothesis again.
Let’s try to refine the search terms in OpenStreetMap search.
Add custom feature: OSM Key place contains “bella”.
As a result, only three places were available. Let’s check them out.
And the first link turns out to be the right one!
I’ll be honest, before I found the place, I couldn’t read the exact name (“bella” or “bolla”) of the restaurant (view first photo). But maybe someone else would have been able to and would have started this investigation with a restaurant instead of a supermarket.
In this case, instead of looking for an signboard that ends in “bka”, you could just look for places that contain the word “Bella” in the title and see all 19 places on Street View.
When there are bits and pieces of at least two signboardsin a photo, it’s just a gift of fate. But more often it happens that there is a piece of only one inscription.
What to do in such a case?
Other custom features (parking, trees, benches, building over 3 storeys high) and other small details could be added to the search. Unfortunately, OpenStreetMap is not very accurate in such small details, but it would be worth a try.
(I’m not sure there are three rows in the picture, I’m just showing you an example.)
Conclusion
Yes, AI geolocation tools rarely determine a perfectly accurate location and they can’t yet replace OSINT specialists But they can help save time when working with traditional tools like OpenStreetMap search (OverPass Turbo).
I’ll be honest, I was very lucky that GeoSpy got the country and city correct. But if it had gotten it wrong, I would have asked ChatGPT and Perplexity to look for company names that end in “bka”.
They might be wrong too, but maybe they would offer some other ideas that would lead me to new hypotheses (e.g. that the combination bk is typical for Polish etc).
