10

Since the war in Ukraine news media often try to geolocate the place where a video or picture was taken by looking at buildings or treelines present in the imagery.

Both Russians and Ukranians often blur treelines or other landmarks when they film their positions on the battlefield. This is to help prevent geolocation.

I understand that a building like a church could easily be geolocated. But how is geolocation using more mundane objects like treelines done? Do they use Google maps in combination with AI?

2

2 Answers 2

11

Logan Williams of Bellingcat gave a talk at FOSS4G 2023 called "Investigating war crimes, animal trafficking, and more with open source geospatial data" where he described how they geolocated images based on trees, power lines, scuffs on the white line in the road and shadows and reflections. The slides will give you an idea of what he was talking about, but there is a video of the talk too.

4

I don't know how the news media do it, but here's how I do it:

  1. Develop a rough idea of where the picture was likely taken.

  2. Look for features that will show up well in aerial photographs. My first choice is gas stations: the canopy over the pumps is easy to spot and unlike anything else in the world. Other things that work well are flaws in road markings or things that cast distinctive shadows.

  3. Scan the target area for the target features. Whenever you find one, check for less-distinctive features to see if they match up.

To take a random Google Street View from a suburb of Denver:

enter image description here

  1. The presence of this travel trailer strongly suggests the presence of a driveway wider than the garage. This is unusual, and makes a good primary search feature.
  2. This basketball hoop casts a distinctive shadow, and the pile of leaves at the base suggests it doesn't get moved much. It makes a good verification feature.
  3. The light patches on the road suggests re-surfacing with a thin layer of asphalt, which has since worn away somewhat. A good primary search feature.
  4. Two adjacent houses with L-shaped ridgelines. Most construction in the area exhibits more variation, so a reasonable primary search feature and a good verification feature.
  5. A wide gravel border on the driveway and walk. A bit small to make a good primary search feature, but rare enough to make an excellent verification feature.
  6. This fence extends to the property line; construction patterns in the area strongly suggest that this means it encloses a back yard. This in turn implies that there's an intersection behind the photographer, which is a good primary search feature.

Two things in the picture that don't generally make good search targets are the roof color (at least in the United States, apparent roof color can vary wildly with sun angle) and the pattern of evergreen vs. deciduous trees (they're easy to distinguish in fall or winter imagery, but most readily-available aerial imagery is from the spring or summer).

1
  • I agree that roof color don't make good features to search but other roof features (like chimneys position) are very useful because they are shown both in land-based and aerial photography.
    – Pere
    Commented Aug 3, 2023 at 15:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.