Wobbly Google satellite images


Clement Valla collects Google satellite photos where the image-processing software went a little wonky and created the illusion of wiggly-wobbly landscapes. It's inceptiontastic!

Postcards from Google Earth, Bridges (via Kottke)

28

  1. Not really a image-processing software issue, and rather how Google Earth maps the satellite images to the recorded height of the landmass in real-time 3D. Since the satellite picture was taken at a fixed angle (and a steep one at that), the rendered result at ground level is akin to a Trompe-l’oeil.

    1. So you’re saying it’s not an artifact of image processing, it’s because of the way Google is processing the image.

      This particular example isn’t just wrong because of an assumed flat image being projected onto a contour, it’s also wrong because the contours don’t reflect the fact that there is a bridge there. In fact, there’s no way to represent the bridge accurately using just a simple elevation model, since there are two relevant elevations: the bridge and the ground under it. An oblique shot could show both.

  2. Exactly as Takashi says – this isn’t a photo stitching problem. The photos have been taken from overhead as a flat representation (as any photo is).

    Google Earth has then laid that flat photo over a contoured surface. When you pan the camera around so that you can see those contours sticking up, the photo appears distorted.

  3. There is a very slight Google Map error / wonkeyness in the train track close to my house in Northern Ireland.

  4. Here is another, different, Google/satellite easter egg I found.

    I have not heard of someone finding one like this before.

    Link

      1. The weird effects with planes on Google thing is new to me. I had not heard of it before. Thanks for sharing that weird image of the two tailed ghost shadowed plane!

        I find it interesting that Clement Valla’s images will probably be there until Google changes their image mapping algorithm. The planes will disappear on the next update of images.

    1. I’m pretty sure I’ve seen planes in Google images before, but not rainbow planes. Spiffy :D

  5. Did anybody bother to check if it doesn’t look like that? I mean, cheap contractors, surrealist architects, earthquakes…..

  6. It may not matter, but is this satellite imaging or aerial photographs? I believe they do use aerial photographs for those really zoomed-in views of your garden/backyard.

  7. This isn’t an error, Google has just discovered that we’re all living in a Matrix-like instance of SimCity 3000. Expect to have your splines reticulated any time now.

  8. These are pretty great. It’d be nice if he included location data. As it is, the filename provides a hint but not an exact location.

  9. Ok.. So I have worked in the photogrammetry and orthophotography industry for the last 12 years so I know a bit about what is going on here.

    First we must understand that the imagery here is areal and not satellite. I know that the button on google maps says satellite but at this scale the imagery is aerial. That Areal imagery is called orthophotography, and the city or county in question paid some company a good chuck of money to fly it and most likely produce a map using photogrammetry (neat stuff, look it up). At the same time, they used the 3D map generated by the photogrammetrist to create a TIN that was then used to create the orthophoto.

    an orthophoto is a stitched aerial image that uses very complex algorithms that take into account the exact point in space where the photo was taken, the lens used to take the photo and the exact contour of the land being photographed to put every pixel exactly where it belongs.

    problems can and do happen when that software encounters a bridge and lots of hands on work is used to fix those problems.

    the problem you see here is what happens when you overlay a good orthophoto over a TIN that is a much higher scale and nobody is there to correct anomalies such as these because this is being don on the fly by the google software.

    1. As a fellow GIS person I concur with ERIKO23. All this seems to be is a high resolution air photo overlay on a low resolution TIN.

      Key is what jetfx said is that a TIN is a “representation”. That is to say it is an interpolated surface based on a finite amount of data.

      So when you take something that is a static image of actual representation of the landscape in this case an air photo, and overlay it, then I am not surprised it looks wonky. In fact it will always look wonky. They could apply a “smoothing” algorithm so that the TIN isn’t so angular, but really it comes down to the TIN is based on a finite low resolution of points in comparison to the photo.

      Mind you, I haven’t worked with this sort of stuff for years, but I always remembered TIN’s as I thought they were pretty cool at the time. I believe TIN’s also regularly involve an “exaggeration” ratio, that heightens differences between peeks and lows to accentuate the differences so you can see them better. So lowering this ratio might also help if it was applied.

  10. A TIN is a triangulated irregular network for those who don’t know. It’s a 3D representation of terrain using lots of small interlocking triangles.

    But ErikO23’s explanation is exactly correct.

  11. I think most of the other images in the gallery are better than the one Cory chose to post. They show the images being overlaid exactly as you’d expect, while still looking wonky. This one is particularly strange because it’s a hugely wide bridge and he’s zoomed in all the way and the terrain exaggeration was probably turned up a bit (in the other ones he’s got it at a “realistic” setting).

    I did geology research that involved extracting data from digital elevation models, and became quite familiar with them – how they’re made and what the pitfalls are. Google Earth usually has excellent DEM data, but it’s heavily interpolated by apparently proprietary algorithms to smooth out properly. If you get the raw data from other sources, it usually doesn’t match Google’s.

    Thing is though that this data is remotely obtained, either from satellites or airplanes, and there are loads of things that cause errors and anomalies. It has a particular problem dealing with deep canyons and so on, so it’s no surprise that there are so many wonky bridge images on google earth, even in relatively shallow man-made freeway canyons like this one!

    You can tell when this data has been hand-tweaked, as in famous places like Niagara Falls (there is an image of the Rainbow Bridge at Niagara Falls in the gallery here – the topography is remarkably realistic, but man-made structures look wonky). A lot of times you can get better data just by looking at topographic maps, which were originally made by people manually interpreting aerial photographs or doing physical surveys and have much better resolution as a result (though they’re difficult to digitize and work with on the computer – a mammoth task in fact for anything non-trivial).

    Anyway, it was a fun game to try to figure out where all these bridges are. As a fan of geography, geology and geomorphology (I am a geologist), architecture, and engineering, I’ve seen a lot of bridges and am familiar with the topography of a lot of places. Most of the famous bridges I figured out, and many of the others I can roughly guess based on what the area looks like. Many of these I recognize as having driven across in person, even some of the non-famous bridges!

    I wouldn’t be surprised if Google Earth starts including 3D models of bridges in the future; they’ve already got 3D buildings so I don’t see why they wouldn’t, at least for famous bridges.

  12. I’m pretty sure ESRI has some new tools in its release of ArcMap 10 that correct some of these problems that are in these snapshots.. My company hasn’t moved to 10 and I don’t work alot with aerial photography but I would expect in a few years this problem will probably go away. But I imagine, a great deal of work will be required.

    1. I hope you’re not a fan of python, because you’re going to have lots of fun with the changes they made from ArcMap 9.3.

      But addressing your main point, I am not aware there is a fix for this issue in ArcMap 10. The issue lies with the data, as most DEMs (digital elevation model for all you non mapping people) strip out the man made structures to show the underlying terrain, and that’s if the data was gathered at a high enough resolution to even pick up structures in any detail. You still have to fix the issue manually, which is why Google doesn’t bother because of time constraints. But it is reasonable to expect that the issue may fixed in the next few years with software smart enough to recognize what it is looking at, and what to do about it.

      @peguinchris
      Almost all of the 3D models in Google Earth are user created. They can be made using Google’s free Sketchup software which is insanely easy to use. After twenty minutes of watching the tutorial videos, I was able to make a model of my house (exterior only) to a very professional looking level of detail. If you’re interested, you can make models of existing structures and submit them to Google Earth. If they meet the criteria, they will appear online in a few weeks.

  13. The rainbow plane is clearly a chemtrail tanker, carrying the chemicals that they disperse into our water supply that cause rainbows in hose-nozzle sprays.

  14. All this technical discussion is moot. The simple and glaringly obvious answer is that Frank Gehry is now being commissioned to design freeways.

  15. I see others have already pointed out that this is due to a two dimensional image taken at various angles mapped onto a three dimensional representation of the Earth’s surface then turned into a three dimensional representation on your two dimensional screen. The three dimensional map does not take into account things like bridges (unless perhaps they are of sufficient height) and you also cannot use the “rotate” function to “look behind” things (though I’ve seen people try nonetheless). Perhaps some day they will allow the rotate feature to cause a different image, taken from a different angle, to load?

Comments are closed.