A Digital Rotating Panoramic Camera
by T.K. Sharpless (TKSharpless@gmail.com)
Rotating panoramic cameras have been around as long as film has. As the camera turns, a slit in the focal plane scans a narrow strip of the image onto the film. In the digital age, a linear CCD replaces the film and slit, and the camera is connected to a computer that records the panorama as a series of one-dimensional digital images.
Desktop document scanners and digital rotating panoramic cameras have a lot in common, so by now quite a few people have made cameras from scanners. The pioneers include professor Andrew Davidhazy and artist Ansen Seale. The basic idea is to mount the linear image sensor and a photographic lens on something that can rotate smoothly and arrange for the motor to turn it. You also have to provide an infrared blocking filter, some substitute for the scanner’s black and white calibration strips; and some software to run the thing.
I’ve been doing that for three years now. What started as an amusing hack has become a full fledged photographic hobby as I discovered what amazing pictures these cameras could make (see my web portfolio).
My first “ScanCam” turned on the shaft of a discarded Makita sander, the rest was mostly wood and baling wire. But it could take sharp 30 megapixel pictures; so I built a better one, then another…. I just finished my fourth camera, based on a 7 year old 600DPI H-P Scanjet 2200c scanner. It is the first one I would rate as suitable for daily use. The full field kit, shown above, weighs 30 pounds including the camera, six lenses, a sturdy tripod and head, an 18 volt NiMH battery pack (in bag under tripod) a laptop computer, a 15 foot USB cable, a special white calibration target, various filters, and a photographer’s backpack to carry it all.
Scancam Mark 4 has comparatively sophisticated mechanics as a result of lessons learned from Mark 2 and Mark 3. The most important ones are 1) the works have to be enclosed against dust, water and static; 2) the base should be small enough to stay out of view with the lens tilted down; 3) there must be no play in the drive mechanism – but it must not jam, either; 4) the optical elements should be rigidly fixed with no adjustments (minor misalignment is best corrected in software).
Like its predecessors. Mark 4 is controlled by a National Semiconductor LM9832, a very flexible scanner-on-a-chip used by several scanner manufacturers in the 1999-2001 era. Because the documentation for the chip is available to the public, I was able to develop special software for scanning photography. This software, by now also fairly sophisticated, works with all of my ScanCams.
The heart of a scanner is a linear CCD image sensor, one pixel wide and thousands of pixels long. The scanner’s lens and mirrors project a narrow strip of the page onto the CCD. In a camera, the CCD sees a narrow vertical strip at the center of the image formed by the lens. The CCD in the Scanjet 2200c has 5,300 pixels 0.007mm wide and the same distance apart; so Mark 4’s sensitive strip is 37.1 mm high by 0.007 mm wide. Actually there are 3 sensitive strips side by side, 0.056 mm apart, covered by red, green, and blue filters. One of the software’s jobs is to align the three colored images.
The core of Mark 4 is a brass tube with a Minolta MD lens mount at the front end and a tall box containing the CCD (which is mounted on the scanner’s main circuit board) at the back.
Inside the optical tube are a round infrared blocking filter, mounted near the lens, and a tall narrow color balancing filter mounted near the CCD. The lens-to-CCD distance, set by machining the brass tube, is 1.7mm more than the standard for the MD mount to allow for refraction in these filters.
The box that houses the scanner’s main circuit board is attached to the back of the optical tube via a round aluminum plate. A shutter, used for dark calibration, slides in a recess in the plate, to cover a narrow slot in the box just in front of the CCD. The shutter is operated by a solenoid connected to the scanner’s lamp power supply, which is under software control.
The CCD box is a U shaped aluminum channel with black plastic end covers and a sliding metal back cover. It has a thick section in front of the CCD, with recesses for sturdy steel posts that support and locate the circuit board. Ribbon cables inside the box bring the scanner’s I/O port and home sensor connections to a connector on the bottom end cover that mates with the external cable.
The optics tube is clamped to a plate that slides in a semicircular yoke for tilting the lens up and down. One side of the yoke has an engraved scale for the lens elevation angle, and there is a bubble level on top to let me set the camera’s rotation axis accurately vertical.
The yoke is clamped to the top of the main shaft, which runs in ball bearings contained in a short tube protruding from the top plate of the cylindrical base. To allow the yoke to rotate, the flexible ribbon cable between the two boards wraps several times around the bearing tube. Right angle folds at each end of the wrapped section let the ends of the cable run vertically A ferrite sleeve attached to the yoke guides the upper end of the cable and helps suppress electrical noise. A plastic disc on top of the base keeps the free loop of the cable from snagging when the lens is tilted up.
The base houses the drive mechanism and the scanner’s power and I/O board. Its bottom plate has a tripod screw socket in the center and holes near the edge for access to the scanner’s USB and DC power connectors.
Mark 4 features a zero-backlash worm gear drive. The primary drive train is mounted on a pivoting carriage, and the worm is held in contact with the gear by light spring pressure. The carriage pivots are in line with its center of gravity, so that the worm won’t lose contact with the gear when the camera tilts or moves.
The primary drive includes the scanner’s stepping motor and main drive gear, which has a toothed belt pulley. A short belt drives the worm gear shaft through another toothed pulley. The motor and main gear are on a section sawed from the scanner’s plastic chassis, which is held between the aluminum side plates of the carriage on two sets of tubular spacers. The worm shaft runs on ball bearings in an aluminum support mounted at the third corner of the carriage.
Those bearings are preloaded to eliminate lengthwise motion of the worm,
The posts that support the carriage pivots are adjustable for height, and one of the pivots is a conical bearing whose clearance is adjusted by a screw. By careful adjustment it is possible to reduce the play in the pivots to near zero while aligning the worm exactly at the center of its gear.
Two leaf springs anchored on the support plate provide the pressure that keeps the gears in contact.
A helical spring housed between the main gear and the top plate provides a torque that takes up any remaining play.
The optical home position sensor, removed from the main circuit board, is inside the drive housing. The moving vane that activates it is mounted on a plastic pulley driven by an o-ring belt from the hub of the main gear. The pulley is about 50% larger than the hub, so that the shaft can turn more than 360 degrees before the vane reaches the sensor again.
A ScanCam differs from a rotating film camera in several respects other than the obvious one that it is operated by software running on a laptop computer. Perhaps most important, it is much slower. The slit in a film camera is around 1 mm wide, while Mark 4’s is 0.007 mm; so the digital camera takes around 140 times longer to capture a given image -- assuming the same photographic sensitivity, which is only about ASA 100. Since Mark 4’s maximum resolution is nearly 18 megapixels per 35mm frame, in practice I take most scans at half resolution, which cuts the scan time by a factor of 4. For example, a 360 degree scan at half resolution in daylight with a 24mm wide angle lens takes 3 minutes and captures 27 million pixels. But I often take 15 minute scans, and occasionally 30 minute ones.
Since the CCD length is slightly more than the width of a 35mm frame, the scanned images are comparable to stitched panoramas made from “portrait format” 35mm photos. A 360 degree scan with a 24mm lens is equivalent to 6 ½ normal photos with no overlap, or about 10 with enough overlap for stitching.
Another very significant difference from a film camera is that the ScanCam needs to be calibrated before scanning. The rather amazing image quality delivered by even cheap scanners is mainly due to the way they correct defects in their sensors and optics. The procedure is as follows. Before each scan, measure the sensor’s raw responses to black and white calibration strips. During the scan, correct each pixel by subtracting its black response, then multiplying by a factor inversely proportional to its white response. The result is a very uniform corrected response, flatter than a film camera’s because the white calibration also removes lens shading; but most important, it corrects for dust and optical imperfections on the CCD.
It is easy to get black calibration data. My ScanCams do that automatically, using an electrically operated shutter to put the CCD in the dark. White calibration is much harder, especially with wide angle lenses. The goal is a uniform white field. That requires manually placing a suitable target in front of the lens. Flat targets such as white cards or ground glass work all right with long lenses, but do not light the periphery correctly with short ones. The best target I have been able to devise is shown in the first photo above. It is a sector of a cone made of white paper, held in a curved aluminum frame. When evenly lit from the side, it puts a very flat white field on the CCD, even with wide angle lenses.
I now do white calibration in two stages. First, in the studio I light the conical target very uniformly, and record each lens’s shading curves at several apertures. The software stores field flattening files that are later reloaded when I set the aperture for taking a picture. Just before scanning I do a “dust” calibration, which removes localized defects (typically due to dust specks) without changing the overall shape of the correction curve. At this stage the target image doesn’t have to be flat, just smooth and featureless – I often use a strip of milky white plastic, or an out of focus card, instead of the cone target.
Both parts of the white field correction are sensitive to lens aperture. The shading correction changes most at large openings and is essentially constant below f/5.6. At small apertures the dust correction is large and varies strongly with aperture; it becomes constant (and often negligible) at large ones. As a practical matter, the dust correction is essential for good outdoor pictures and often unnecessary for indoor ones since the field flattening curves also correct intrinsic pixel-to-pixel sensitivity differences.
An advantage over most rotating film cameras is that the ScanCam takes interchangeable lenses, so can make panoramas in a wide range of formats from near-spherical to very long and thin, as well as high resolution rectangular images. I use 6 lenses on Mark 4, ranging from an 8mm fisheye (image height 180 degrees, but only 3700 pixels) to a 55mm portrait lens (image height 37 degrees, 5300 pixels). I mostly use a 16mm fisheye (image height 153 degrees, 5300 pixels) for “superwide” and “supertall” architectural pictures and the 55mm for classic panoramic landscapes.
Having lots of pixels and a wide choice of angular resolutions makes ScanCam pictures natural raw material for reworking on the computer. I use the Dersch PanoTools and some programs of my own based on Dersch’s source code to adjust viewing perspectives, sometimes radically and sometimes subtly, to make images better represent their subjects as I (would like to) see them. I sometimes stitch multiple scans into even larger images, and sometimes combine different versions of a scene.
One of the few things that film can do better than CCDs is tolerate local overexposure. Camera CCDs have expensive “antiblooming” circuits that keep the excess charge from badly overexposed pixels from damaging too many of their properly exposed neighbors. But even then the spoiled area is typically larger than it would be on a good film. Worse, the damage takes the form of long streaks of whited-out pixels rather than compact blobs. Unfortunately, scanner CCDs don’t have antiblooming circuits at all, so that even one pixel overexposed by 2 stops or more spoils a whole column of the image. That makes ScanCams useless for high dynamic range imaging, and hard to take normal pictures with as well. Any well lit scene is apt to contain a few bright reflections, perhaps quite small, that will throw nasty streaks across a ScanCam image; and windows regularly wreck interior shots. I have to be extremely careful to avoid that sort of thing, and often have to underexpose the parts I’m interested in just to avoid having a bit of the sky blow them away altogether. Of course motion and unsteady lighting also spoil ScanCam images (though some kinds of motion artifacts are amusing or even beneficial) so I have to be very careful about the subject and the weather as well.
The bottom line is that this is a very deliberate kind of photography, that puts a premium on careful observation and planning as well as luck. But maybe that’s true for other modes of panoramic photography, too.