Fast Generation of 3D City Models

FleaPlus writes "New Scientist and UC Berkeley News report on research by Avideh Zakhor and Christian Früh on rapidly and automatically generating photo-realistic 3D models (coral cache) of city environments. This is reminiscent of research mentioned here in March, although the approach is rather different. Applications for areas like urban planning, entertainment, emergency response, and the military are discussed. Zakhor's group is also working on 4D systems, to allow for recording of events over time as moving 3D models."

SGI?

[dclaw]

If I remember correctly... companies like Silicon Graphics have been doing this for several years for the military using their reality center systems to analyze satellite photographs of combat zones in real time.

Re:Photo Realistic 3d?

[Anonymous Coward]

What's new here is that street-level capture took 25 minutes of driving and was completely automated.

lidar

[kebes]

This technique should be replaced with (or combined with) LIDAR techniques. LIDAR (Light Detecting And Ranging) is like a laser version of radar, and it can be used to generate 3D maps of objects or topography fairly quickly. A LIDAR-enabled plane can do a fly-over of a city and quickly generate 3D maps. For instance, check out these LIDAR images [loc.gov] of New York pre and post 9/11.

I've always thought that these 3D maps of cities should be combined with satellite imagery so that you could 'paint' the rooftops onto the buildings and streets. In fact, most satellite images are at a slight angle, allowing you to paint the facades of buildings. The techniques from TFA show how you could paint even more facades in the 3D model/map fairly easily. I think the LIDAR data would greatly simplify the painting algorithm, since the correct 3D shape of every object is known.

I can't wait for the day when we can browse 3D maps online of any city. It would be useful for surveyors and architects of course, but also useful for planning trips, learning geography, and so on.

4d Futurists

[sellin'papes]

Reminds me the futurist painters in the 1920s and 30s. They would try and capture an object from all different angles and movements for a period of time in one image.

For instance Picasso (in his early days) did paintings of peoples faces from different angles and with different emotions all as one face.

Re:SGI?

[Janitha]

But that was mostly done with scanning the distance around a given point (or several points), and it had no image but just a depth map, even if there was image it would be used to simply identify more the object at a given point, not to this scale.

This is quite impressive, yet they still have to work out those glitches that are quite visible. Its fast, but they are giving away precision for speed.

The other thing I don't understand is the used for rapid generation of these, speed is good, but cutting down on how accurate it is for speed is not acceptable. A better view would be pre-rendering the images with more precision/accuracy. Not like buildings come and go every day.

Wonder how it fares with European cities

[Matthias Wiesmann]

All the examples shown in linked article are US style cities, with straight roads and relatively simple roughly cubical buildings. I wonder how this technique fare with more complex buildings and street arrangement, as you find in the center of many European cities.

I would really be impressed if they could automatically generate a 3D model of the three dimensional mess that is Lausanne [lausanne-webcam2.rcv.ch]: