heres the proper site for it
www.tpd.tno.nl...
The photo
To display the photo a Flash viewer SWF control is used which requires Macromedia Flash Player v6 (which is a free download). Please check your
security restrictions, since some computers might block the installation/usage of third party software. If this happens, reduce your security
restrictions temporarily.
Without a broadband connection or if there is a lot of traffic on the server, the online viewer might be a little slow. Such a large amount of data is
involved that we had to make a trade-off between image quality and transmission speed. To keep the transmission speed reasonable, we used JPEG
compression on the image data and accepted some loss of resolution at maximum zoom.
Use the toolbar buttons below the image to zoom in and out or move around. Click on the image to centre the region of interest and move the slider to
change the zoom level.
The approach
The 2.5 gigapixel image is a composition of images rather than a single image. TNO developed a sophisticated approach to merge the many images - all
600 of them. The photos were taken automatically using a modern consumer camera and a powerful 400 mm lens. The camera was positioned automatically
using a computer-controlled pan-tilt unit. Each of the 600 photos overlaps, an arrangement that ensured very accurate positioning and enabled us to
stitch the images automatically into one giant image of 78,797 by 31,565 pixels. The most difficult tasks were processing these large images and
comparing the overlapping images.
Long 'exposure time'
All 600 photos were taken over a period of 1 hour and 15 minutes. Taking a single photo and moving the camera to a new position took approximately
seven seconds. Thanks to the long 'exposure time', some interesting artefacts are visible at the edges of the various photos. They include a parked
car that seems to merge into a bus and a walking torso.
After the photographic session, considerable effort was required to calculate the final result. It took about 24 hours to compare the overlapping
photos and optimise them. Stitching the photos into one image required the capacity of 5 high-end pcs for three full days.
The final result allows the viewer to zoom in on the city of Delft and its surroundings at a resolution never seen before.
Technical details
Below are some facts and figures about the gigapixel image:
Final image dimensions: 78.797 x 31.565 pixels
Number of pixels in final image: 2,487,227,305 (2.5 gigapixel)
Final image file format: 24-bit colour bitmap
Final image file size: 7.5 GBytes
Number of source images: 600
Number of pixels in source images: 3,537,408,000 (600 images * 3008*1960)
Lens focal length: 400 mm (equivalent to 600 mm on a 35 mm camera)
Aperture: F22, Shutter speed: 1/100, ISO: 125
Horizontal field of view of final image: 93 degrees
Time required to capture component images: 1 hour and 12 minutes
Time required to match overlapping images: 20 hours
Time required to optimise project: 4 hours
Time required to compose the image: 3 full days using 5 high-end pcs
Time required to blend seams / correct misalignments / finalise image: 2 days
Detailed approach
General strategy
One of the first steps was to decide how to take the photo. Given the 2.5 gigapixel target and the 6 megapixel resolution of many digital consumer
cameras, it was apparent from the start that hundreds of photos would have to be taken and processed. From day one it was decided, therefore, that the
only option was a fully automatic shooting procedure, i.e. moving the camera automatically using a motorised and computer-controlled pan-and-tilt
unit.
Location
The location had to satisfy constraints like attractive view, sufficient height and easy accessibility. Since such natural locations are rare in the
Netherlands, the Electrical Engineering department building on the Delft University of Technology campus was chosen. This building provides a clear
field of view from a height of almost 100 m. No nearby buildings obstruct the view.
location of shooting
File formats
After some initial experiments it became clear that existing panorama stitching tools and file formats were not capable of handling multi-gigapixel
images. A simple example is the TIF file format, which only supports files smaller than 4 GBytes due to the 32-bit offsets used in its tables. An
uncompressed 2 gigapixel photo is at least 6 Gbytes (3 bytes per pixel) and therefore too big for the TIF file format. Together with the definition of
a new, large-image file format, many existing software tools and scripts had to be rewritten.
Image storage during shooting
Taking hundreds of photos in RAW format (necessary for later processing) in a single run requires gigabytes of storage space in the camera. The option
to change memory cards during shooting was rejected as this would lead to disturbances of the motion platform while swapping the cards. Another option
was to use a single high-capacity memory card (8 GBytes cards are now available) but this requires FAT32 support from the camera). In the end, it was
decided to use a high-speed FireWire link between camera and computer and store the photos on a laptop computer. Camera and lens The camera used was a
Nikon D1x. It provides 6 megapixel resolution and a FireWire interface with remote-control capabilities. This digital SLR camera needed a long
telephoto lens to provide enough detail in the final image and to prevent us ending up with an extremely wide field of view. The lens used was a Nikon
AF VR Nikkor 80-400 mm f/4.5-5.6D ED. As the digital SLR camera has a 1.5x magnification factor, this lens compares to a 600 mm lens on a 'normal'
35mm SLR camera. The lens comes with the option of vibration reduction technology but this technology was not used. During tests the vibration
reduction did not improve image quality when the camera was mounted on a tripod.
Motion platform
The motion platform consisted of a pan-and-tilt head to rotate the camera around two axes (left-right, up-down), and a motion controller to control
the motors on the pan-and-tilt head. A Dspace system was used to control the motors while the photo was being taken. While this system offers rapid
development, it is voluminous, very expensive and requires additional computers. The dSpace system was replaced recently with a compact and much
cheaper 2-axis motor controller from Galil.
Tools and equipment
Below is a complete list of the hardware and software tools and equipment used during the making of the 2.5 gigapixel photo.
Hardware components:
Camera: Nikon D1x
Lens: Nikon AF VR Nikkor 80-400mm f/4.5-5.6D ED
Tripod: Manfrotto 755B
Pan and tilt head: Camera Turret Company Digital PT20 head
Motors, gear-box, encoders and amplifiers: Maxon
Power supply: Traco Power TIS 150-124
Motion controller: dSpace modular hardware, later replaced by Galil DMC-1425
Computers: two standard laptops and various desktop computers
Software:
Remote camera control and image conversion: Nikon Capture
Motion control: dSpace Real-Time Interface together with MathWorks Matlab and MathWorks Simulink
Motion control: Galil DMCWIN API
Control point generation: Autopano
Panorama assembler: PTAssembler
Lens deformations: Panorama Tools 2.6.ML12
Speed improvement: Faster Pano12.dll
Blending: Enblend
Scripting: MathWorks Matlab
On-line viewer: Zoomify
Several integrating software modules for camera control, motion platform control and overall control developed by TNO.