I've written about Gigapan before - a project being carried out at CMU which has adapted photostitching technology originally developed for exploring Mars rover images to consumer facing applications. There are a number of recent updates to the project including the announcement that the robotic camera used to create the photographs is available in limited numbers:
In conjunction with Charmed Labs, we've just announced a public beta for the GigaPan robotic camera mount. Here is the Carnegie Mellon press release regarding the announcement on September 26. This first model is capable of capturing multi-gigapixel, explorable panoramas with most compact digital cameras. Introductory pricing for the beta device is $279. We're looking for people to help! If you're interested and have time to capture lots of panoramas in the next couple of months, please apply here no later than October 19. Applicants will be informed by October 26.
I've been familiar with the project for a while, and Wakako had the opportunity to use the robotic camera platform on our recent trip to Vancouver. An image taken there is available on the Gigapan site.
The Gigapan site has some great features, including the ability to launch some of the pictures in Google Earth.The site has (right now) 762 images available for zooming, panning and exploration. In addition, there is a map which allows you to explore the available images by location (note that not all of the images are geocoded).
Interestingly, despite the name of the project, most of the images hosted there use only a fraction of a gigapixel. Each image is described by a number of features including the number of pixels. The highest I've seen so far is 2.81.
The project as a whole has a number of key technologies:
- Robotic panorama control for cameras: plunk in a consumer grade camera and the platform rotates, taking pictures as it goes.
- Image stitching software: once all the images are taken, they need to be stitched together to form the panorama.
- Image server and client software: provide a smooth image exploration experience.
One of the challenges with the stitching of images is the fact that the scene can change between the time of each image capture and when this occurs on a seam, one gets artifacts like the following:
It would be interesting to see if this type of problem could be completely solved by using seam carving as demonstrated by Shai Avidan and Ariel Shamir.