Join 3,422 readers in helping fund MetaFilter (Hide)


Behold, the zoomable panorma
October 2, 2007 4:55 AM   Subscribe

Science Daily reports that researchers at Carnegie Mellon University, in collaboration with scientists at NASA's Ames Research Center, have built a low-cost robotic device that enables any digital camera to produce breathtaking panormamic images called Gigapans.
posted by Dave Faris (25 comments total) 5 users marked this as a favorite

 
Cool!
posted by phrontist at 5:02 AM on October 2, 2007


You can also do this with AutoStitch and video from any camera. Just pan slowly in all directions, export every 10th or so frame, and throw the still frames at the program.
posted by RobotVoodooPower at 5:24 AM on October 2, 2007


Really awesome, but wouldn't a better name be GigaZoom?
posted by DU at 5:29 AM on October 2, 2007


RobotVoodooPower, doesn't it look better if you have a pano head?
posted by chuckdarwin at 5:40 AM on October 2, 2007


Also, the actual device in question.
posted by DU at 5:41 AM on October 2, 2007


I'm not that impressed by the end result, it's easy enough in PhotoShop CS3 to stitch photos together to make one big high resolution image. But I wonder what the robotic camera mount adds to the process... I assume it makes the stitching process automated? That would be cool.
posted by Mcable at 5:49 AM on October 2, 2007


The image that loaded up for me first was this one, and if you look at the woman in the front row of the central group, 4 seats from the right side (her left) she appears to have horror-movie esq dual head, as does the woman two spots to her left (our right). The second woman isn't as noticeable without the two big pairs of sunglasses.

But yeah, this technology is hardly new.
posted by delmoi at 5:53 AM on October 2, 2007


this gritty urban scene is pretty cool, though.
posted by delmoi at 6:00 AM on October 2, 2007


I'm not aware of any device like this. It can take 100s of shots. Previously most pano was 4 or 8 pictures either manually with good dexterity or some kind of tripod with multiple cameras. What makes this different and new is that it can take 100s of photos and thus allow for deep zooming detail. Deep zooming was possible before but it required a super high-rez camera. Anyway, all these features with a consumer grade device is a real breakthrough.
posted by stbalbach at 6:03 AM on October 2, 2007


some kind of tripod: Pano head
posted by chuckdarwin at 6:15 AM on October 2, 2007


I originally read that as "Gigapants."

I was a little disappointed I was wrong.

But this is cool, even without the pants.
posted by papercake at 6:35 AM on October 2, 2007


I'm familiar with autostitch, RobotVoodooPower, and I know that it's really tricky to get all areas of the shot. Granted, a robot that takes all of the pictures on top of a program that stitches them all together sort of takes the photographer almost completely out of the equation... nevertheless, I've signed up to be a beta tester for the device.
posted by Dave Faris at 7:07 AM on October 2, 2007


delmoi: "this gritty urban scene is pretty cool, though."

For a little Pittsburgh easter egg, zoom in on the two big smokestacks at the center of the picture. Oh and the Strip's not really that gritty, that's all show for the tourists.
posted by octothorpe at 7:29 AM on October 2, 2007


Just in reply to this and forgive me if this has already been posted, but this is amazing... 360 panoramic video mash-up with google maps. Japanese predictably.
Roll over a point marked "VM".
Now click on a blue directional arrow on the map.
The map will show exactly where you are and which direction you are pointing.

Produced with this camera .
Samples of footage it captures from a fixed point.
posted by 4candles at 7:30 AM on October 2, 2007


Thing doesn't seem to work for me. After I allow scripts and Flash, I just get a black rectangle with some controls on the left side that don't seem to do anything.
posted by Kadin2048 at 8:38 AM on October 2, 2007


This is really cool, thanks DF.
posted by jonson at 9:03 AM on October 2, 2007


Neat! So where can we buy one of those?
posted by reformedjerk at 9:26 AM on October 2, 2007


the Strip's not really that gritty

Too bad there isn't one from when we would explore there as kids in the 70's. It was gritty then. I still wonder if the guys who paid us $10 to wiggle though a little window and unlock the door had really lost their keys.

One of the key functions of a device like this is to move the camera around the true optical center, which is offset from the normal tripod mount and varies for each lens.
posted by StickyCarpet at 9:27 AM on October 2, 2007


I'm looking at this one and being amazed at how easily you can make out the license plate numbers.

Prediction: In 10 years, this technology will replace ubiquitous police cameras. There will be at least one on every street corner, with hardly any "bubbles" of non-coverage. All the data will be permanently stored on massive HDs, and will be recalled when the police believe there has been a crime in the camera's vicinity.

Kinda like Minority Report, but without the causality violations.
posted by Avenger at 10:36 AM on October 2, 2007


papercake : But this is cool, even without the pants.

I can't tell you how many times I've used that line.
posted by quin at 1:22 PM on October 2, 2007


As a seriously amateur photo-lay-person, this is awesome because you can set the exposure and f-stop to some crazy settings, since you're not holding the camera. Everything can be in focus. Very nice.
posted by id at 2:14 PM on October 2, 2007


Well, I finally got the site to work. (It didn't play nicely with FlashBlock or NoScript, apparently.)

It's neat, but they don't seem that terribly different from what you can do with a regular digital camera and Autostitch or Hugin. It's just that when most people make panoramas out of 10 or 12 (or more) 6MP images, they scale the final product down to a reasonable size rather than leaving it at 72MP, because there's an assumption made that the detail is unnecessary when you want people to appreciate the entire image.

The problem with big, very-high-res panos is viewing them. In the real world, people can get closer or further away, but it's more difficult to do that on a computer, and most sites like Flickr don't support zooming very easily. So as a photographer, you drop the resolution down so that you can fit the whole thing on a computer monitor without scrolling, if the screen is your output device.

So the thing that makes these practical is the viewer, with it's little scroll control and live-downloading of the additional (high-res) portions. That's the real key.

Until you have an interface for dealing with lots of data and presenting it to the viewer in a cohesive manner, collecting the data doesn't make much sense. If more web sites provide ways of sharing very-high-res panoramas, people will start taking more of them, just using the tools that have always been available (for a few years at least).
posted by Kadin2048 at 5:08 PM on October 2, 2007


Slightly off-topic: anyone here know anything more about Haeberli's panoramic compositing process?

Haeberli is this cool graphics guy who used to be at SGI. He has an excellent computer graphics site called Grafica Obscura. One of the projects on that site is Automatic Panoramic Image Merging. I've been looking for more information on that process for years, but no source code is available -- I believe you can get executables for an ancient version of Irix, but nothing recent.

(Haeberli has done lots of really cool graphics-related things, the best of which is probably Lamina.)
posted by phliar at 5:13 PM on October 2, 2007


@phliar: I don't know about his software specifically, but there is an OSS package called "Hugin" which is built up mostly from some non-GUI tools called (unsurprisingly) "Panorama Tools" which does panoramas (and some other interesting transformations besides). You can download and play with it in an afternoon.

I think that most panorama software works according to the same basic principles, some is just more automated than others. Hugin lets you control the level of automation and see the process as it goes through the steps. The really important part is taking the images that overlap and defining "key points," points in each image that represent the same point in realspace. With enough overlapping key points defined, the software can merge the images together, rotating, keystoning, and otherwise distorting them to compensate for camera motion (pan, tilt, zoom, rotation, and some weird lens barrel factors are the four variables I think, plus your X and Y offset for each image).

Programs like Autostitch just do the keypoints business all automatically, and go from a set of images to a full pano in one operation. (Hugin can do this with plugins.)

Here's a list of tutorials
; just reading some of them will give you an idea of the process. Even if you never go on to use the software, you'll get an understanding of what's being done behind the scenes.
posted by Kadin2048 at 12:04 AM on October 3, 2007


I believe that Haeberli's approach is more than the run-of-the-mill panoramic compositors (and I've looked at most of them). This is how Haeberli describes it:
If you've every tried to make a panorama from a series of photographs, you may have noticed that it's very hard to make the pictures overlap perfectly. This is due to the fact that the film plane gets tilted as the camera is pointed in different directions. Unfortunately, it's not enough to translate, rotate or scale the images to make them merge. You must apply a projective transformation in order for the images to seamlessly overlap.
Note that Haeberli's approach does not require you to specify control points etc.
This image was created without any user interaction. The only input was the set of 13 source images and about 30 minutes of CPU time. To do this, first the program warped and correlated each image to all the remaining images. For each pair of images, a difference value was found to describe the quality of the overlap. A directed graph was then created where each node represented one image, and the distance values were used to create bidirectional edges between the nodes. Finally, the graph was analyzed and the best overlaps were used to create the final composite image.
Now that's cool!
posted by phliar at 1:55 PM on October 3, 2007


« Older In the U.S., motorists do not pay their way....  |  The 12 Memorable Moments From... Newer »


This thread has been archived and is closed to new comments