14 replies to this topic

[Matthew Hampton]

What with all the hubbub of the new Google Earth browser plug-in - I thought an interesting link pertaining to 3D landscapes and browser functionality ought to be investigated.

Here is a link to a 3D model of Stockholm. The technology came from the Saab Labs. They offer a uniquely different java-based solution that differs from both Microsoft's VirtualEarth and the GoogleEarthBrowser (and it currently works on Linux/Mac).

[Lui]

Well this puts Google Earth and VirtualEarth in 3D presentation far behind. I think that using LIDAR data and oblique and aerial photos results can be even better.

[DaveB]

Luta Karta made me think of lutefisk. "How to Lye with Maps?" <groan>

[frax]

Too I can't run it. It crashes IE, and in Firefox the plugin complains that the java drivers are not up to date... Maybe it is because I don't have powerful 3d capabilities in my laptop.

Note that this is not Saab the car manufacturer (owned by GM) but the defense contractor (part of Sweden's military-industrial complex).

[ELeFevre]

It' won't work for me either.

[Rick Dey]

Very Impressive
Works fine for me in Safari 3.1.1
That's nice since the Google Earth API isn't Mac compatible.

[A. Fenix]

i'm also having no problems (other than slow download speeds) using Firefox 2.

i would love to look "under the hood" to see how they were able to drape the imagery over lidar.

very impressive!

[Lui]

i'm also having no problems (other than slow download speeds) using Firefox 2.

i would love to look "under the hood" to see how they were able to drape the imagery over lidar.

very impressive!

I don't think that they are using LIDAR for elevation data. My guess would be that they are using automated surface TIN derived from photogrammetry stereo pairs. Elevation points in TIN seems poorly placed (on roof edges,...). With good LIDAR (towns are usually scanned with more than 4 pt/m2) elevation points should present the surface better. Well my generalized surfaces TINs are looking more dense and they defines edges better. The other possibillity is that the 3D engine TIN points limit is low so they have to generalize surface more.
"How they were able to drape the imagery over lidar?" Now this is the most interesting question and I hope that someone can anwser it.

[Matthew Hampton]

I'm not sure that LIDAR is part of the picture. <groan>

<<< From C3>>>
The technology is based on high-resolution aerial photography with carefully calibrated cameras. For every picture, the camera’s position and angle are calculated with extremely high precision, using a very advanced navigation system. This is what enables us to give each pixel its geographical position with decimetre accuracy. Then, using stereovision technology, we combine two sequential pictures to measure the area’s height profile.
"The process is fully automated, producing the following completely without manual work:
* Longitude and latitude of every pixel.
* Height of every pixel.
* A terrain model, by choosing the best pixels from a sequence of pictures.
* A complete, ready-to-deliver 3D model saved as a database."
>>>>>>>>>>>>>>>>>>>

Regading draping LIDAR: I suppose you could find the the closest LIDAR posting to a pixel centroid and position that pixel based on the XYZ, but what about angle? Since available imagery is much coarser than LIDAR postings there's inherently going to be some generalization. I have seen LIDAR point clouds that are colored based on RGB values that look pretty neat, but you would certainly have great difficulty serving-up a whole interactive city!

[Hans van der Maarel]

Regading draping LIDAR: I suppose you could find the the closest LIDAR posting to a pixel centroid and position that pixel based on the XYZ, but what about angle? Since available imagery is much coarser than LIDAR postings there's inherently going to be some generalization. I have seen LIDAR point clouds that are colored based on RGB values that look pretty neat, but you would certainly have great difficulty serving-up a whole interactive city!

I've been heavily involved with Virtual Apeldoorn, so I've seen my fair share of 3D stuff, but this is great. Loads pretty quickly, no need to install a plug-in (FF3 on Windows here, no problem).

The quality seems to go down a bit once you get into the details. The area around Slussen (Hugo will know where that is) seems... well... odd...

I really wonder what would be the end result of using this process for the building combined with a more general DEM for the terrain...

[frax]

Slussen is an interesting place - it is a traffic carousel that was built in the fifties, and it is now falling apart. The plan is to tear it down within a few years and replace it with something more suitable. A minority of people argue that it should be kept - but renovated - for the historical/architectural/cultural reasons, but it looks like that is not really realistic.

It was designed for left-hand traffic (which we had until 1967) and dimensioned for a much lighter load of traffic (both in volume, and weight of trucks).

"Slussen" actually means "the locks" and a part of the whole set up is the locks between lake Mälaren and the Baltic Sea (only for smaller recreational boats). There is also train, subway and a bus terminal. And a lot of odd passageways that you may want to avoid after dark (and they all smell of pee...). And Stockholm's best music venue - Debaser - is hidden in there as well.

[Lui]

Thank you to describe a technology behind this product. It really seems like a digital photogrammetry approach with a good image orientation. Today it is not a problem to get under decimeter accuracy of the flightline. Using LIDAR scanner and camera at the same time enable us to correct IMU orientation to get a high precision orientation angles. We are usually talk about 2-5cm flightline accuracy and less than 8cm of LIDAR accuracy and around 1 pixel accuracy in semi resolution orthophoto (10cm). Producing an orthophoto without collecting tie points from this kind of flightline and on-site calibrated camera and system can be done.

Regading draping LIDAR: I suppose you could find the the closest LIDAR posting to a pixel centroid and position that pixel based on the XYZ, but what about angle? Since available imagery is much coarser than LIDAR postings there's inherently going to be some generalization. I have seen LIDAR point clouds that are colored based on RGB values that look pretty neat, but you would certainly have great difficulty serving-up a whole interactive city!

I didn't mean to use whole LIDAR point cloud and "project" RGB values from images onto points. The result will be heavly depended from point density that is in a case of LIDAR unhomogeneous. I would use just significant LIDAR points (key points) that form a generalized surface (TIN) model and drape a photo onto it. The reduction of points is significant usually from 1:20 - 1:1000 scale. Then a simple LOD would be fine to reduce triangle count. Maybe moving bubble(s) approach is enough. But the real problem is how to drape images onto vertical faces, especially street level images. The precision of image orientation, selection of the most perpendicular pixel to surface and high density LIDAR (to collect more points near edges) is crucial here. And we should not forget a color adjustment problem. Using both aerial and ground based imagery can be tricky.

But the really power is hidden in combination of both approaches. Even using high density LIDAR data (>50 pt/m2) we can't extract edges at planimetric accuracy that a stereography can provide (laser doesn't reflect from an edge). Imagine that we have LIDAR point cloud and stereographically extracted point cloud. Let assume that at edges the image point cloud provide better accuracy. In this case nearby LIDAR keypoints can be moved in position where image cloud points are.

[frax]

Hans - when you mentioned Slussen - one really interesting thing there is "Katarina hissen" which the thingie got totally messed up, in an interesting way. If you look above the Slussen area you see what looks like a narrow building with a totally weird facade - a car and a sidewalk zebra crossing smudge out on it. This is actually a bridge with a one story gondola hanging underneath it.

[Matthew Hampton]

That's pretty funny drape on the elevator...

[Matthew Hampton]

I am giving this post a bump (and an addendum to the title) given the recent sale to a western company. The company has not yet been named, but rumored to be Apple, but it could also be Google, Microsoft, or ESRI.

Followers of browsable 3d mapping should be a bit excited as the implications seem pretty positive.