Digital Imaging SIG - SVCS

While poring over aerial photos of a coastal valley in southern Peru, I noticed the fuzzy images of several large, complex, grid-planned sites in a virtually unknown region. Visiting the sites, it was easy to see that the architecture was complicated, but difficult to understand the layout from the ground. Some aspects of the plan and walls suggested that the sites might have been built by the Wari, a pre-Inka expansive state dating to around AD 500 to 1000, but a millennium or so of debris flows from the valley walls and a layer of volcanic ash from AD 1600 cover the surface, hiding the artifacts that would normally help to date the sites. Understanding the architecture will be key to guiding excavations in three of these sites this summer, and to accurately mapping them using a sophisticated GPS and laser rangefinder system.

But how to get a good, detailed overview of the sites? Satellite photos available for research top out around 15 meters per pixel, nowhere near enough resolution to see the layout of the sites. Resolutions down to a few meters are available if one's budget is large, but even that would not be good enough for walls that are typically 40 cm wide, around rooms and storage spaces often as little as a meter or two across. (See ground level photo at left. Click on any photo for a full size view)

Stereo air photos shot from the 1940s through 1970s for topographic mapping show much more detail. 600 dpi flatbed scans of contact prints of these air photos yield a resolution of roughly 70 cm per pixel. This reveals the general layout of the major courtyard divisions, but is still too fuzzy to show smaller rooms, much less doorways, alleys, verandas, and other features. Expensive 2032 dpi drum scans of the original negatives yielded only high-resolution images of the grain in the film, equivalent to roughly 60 cm per grain.

The solution proved to be flying a digital camera at a lower elevation, using a kite. Alternatives such as suspending thecamera from a balloon were rejected because getting to the sites requires crossing a river and hiking a few kilometers, which would make carrying tanks of gas impractical. I was also concerned about losing the balloon and all theequipment attached to it if the string broke or got away in the strong wind. Solutions involving radio-controlledairplanes or helicopters were rejected as too fragile for the rough and dirty field environment and the often strongwinds, and would have been both expensive to buy and difficult to transport.By using a large parasail kite with no spars (a Sutton Flowform 16), all the equipment needed fits easily in a daypack, with room left over for lunch and other necessities.

The results, based on stitching together many pictures shot straight down from different positions over the site, are spectacular. Resolution is roughly 6 cm per pixel, clearly showing individual rocks in the walls. Even higher resolutions are possible by flying the camera lower, or using a less wide-angle lens.

 

 

 

The camera is suspended from the kite string, not the kite itself. One launches the kite, and once it is flying at a low elevation, the camera is attached to the line. The kite raises it as the line is played out. The camera hangs from a "Picavet" suspension, named after Pierre Picavet, who invented it in 1912 for this very purpose in military applications. It is an arrangement of strings and pulleys or eyelets, attached to the kite line at two points by wire clips, which automatically tends to level itself as the angle of the kite line changes with the wind. Below this suspension hangs a bracket or box holding the camera.

There are many hobbyists practicing kite aerial photography, and almost as many designs for these camera mounts. Many examples can be found on the web by Googling "kite aerial photography" or KAP. One excellent starting point is www.kaper.us. While some designs allow the user to remotely aim the camera, changing the tilt and direction of view, some even with a video feed for aiming, I opted for a simpler, cheaper, and more durable approach that would protect the camera from crash-landings and ubiquitous windblown sandy grit. My camera mount completely encloses a simple point-and-shoot digital camera in a lightweight metal box with a skylight filter window for the lens, and a combination sunshade and crash-landing shock absorber. While the box can be set in any orientation by adjusting a few wingnuts, it cannot be aimed remotely, so there are few delicate moving parts. For my needs, I generally point the camera straight down. The shutter is tripped by a servo motor normally used for radio-controlled airplanes. The motor has a short arm that simply presses down on the camera's shutter button to take a picture. For cameras with infrared remotes, a cheaper and lighter-weight solution involves mounting an infrared LED in front of the sensor window on the camera, and controlling the LED with a simple radio remote, which some users rig from garage-door openers. Infrared remotes designed for consumer cameras don't have enough range for KAP. Cameras that accept professional radio remotes are generally too expensive to risk dropping from a high-flying kite.

The entire setup, including the cost of the kite, line, and associated parts ($150), camera, memory, and rechargeable batteries ($290 last year, less now), remote control airplane servo, receiver, transmitter, and rechargeable batteries ($330), and the parts for making the camera mount, suspension, and carrying case ($75), totaled about $850. Since I already had the 4 megapixel point-and-shoot camera setup, and I had a friend willing to donate some unneeded model airplane gear, my actual investment was under two hundred dollars.

Does it work? Well, the kite is rarely completely still in the air, and despite the Picavet suspension, the camera does swing around. Many shots are blurred by motion. An ideal camera for KAP would have a large aperture lens and/or high ISO sensor, with correspondingly high shutter speeds. However, by taking six or eight exposures at every location, I found that I almost always got at least one useable image with my ordinary point-and-shoot Olympus D-580. The beauty of digital photography is that there is no cost to this approach, which would be prohibitively expensive and time-wasting with a film camera. The only limitation is the memory in the camera -- so it is wise to count the number of exposures in order to not exceed the capacity of the camera's memory card.

Often, a single KAP shot is interesting or graphically appealing on its own. For my purposes, I needed to stitch together many images to make photographic pseudo-maps of the three archaeological sites. To do this, I use Hugin, a remarkable public-domain front end for the extraordinary optical correction and stitching Panotools software by German mathematician Helmut Dersch. Both can be downloaded, legally and for free, from hugin.sourceforge.net. This package, designed for stitching multiple rows of images together for panoramas and a range of virtual reality applications, will geometrically adjust images for lens distortions like pincushioning or barrel distortion, match different scales, rotations, angles of view, and so on, even correcting exposure differences that may cause the corners of images to be darker than the center and normalizing the colors and exposure of all the images. This flexibility and power is necessary for assembling KAP photos shot by a camera swinging around in the air and moving from place to place to cover a large site. It works even better for stitching hand-held panoramas and the like.

The assembly process begins with selecting the images to be stitched, a task only a jigsaw puzzle addict could love. In my case, I typically select forty to seventy-five images. Since most of these have soft focus anyway, I lose little image quality by downsampling them substantially before stitching them into what will still be an enormous image. After dragging and dropping the images into hugin, the main task is to select matching points from every pair of images that overlap. For a series of images taken from a single point, as one would normally do for a panorama, the software can automatically pick corresponding points, but KAP shots vary so much in scale, rotation, and geometric distortion that this feature does not work well for them. To select points manually, Hugin displays any two images side by side, and the user clicks on an identifiable point on one image, and then the corresponding point in the other. The software helps by using pattern-matching to fine-tune the points, displaying the relevant sections of each image at full resolution. One can override the suggested placement if the software does not get it right, as occasionally happens with images at very different scales or with differing distortions. I generally pick five or six corresponding points per overlap. More would produce even better results, but this works well enough. With forty to seventy images, each overlapping with numerous others at different scales and rotations, it takes a long time to select all the points.

To control the geometry of the stitched output, we placed visible markers in a straight line across each site at 25 meter intervals. These can be identified in Hugin as points that should fall on a horizontal or vertical line in the output. My results would have been more geometrically correct had I placed a grid of control points on the site and then defined numerous vertical and horizontal alignments for the output file.

Hugin can output the stitched results in many different formats. I choose the Photoshop format with layers and cropping masks. This results in every image being adjusted to match the others, but remaining a separate object on its own layer. I then open the file in an image editor (I use Corel Photopaint, but Photoshop would do the same tasks). I can cause the best image for any given point in the montage to show by changing the stacking order of the layers, and by adjusting the cropping mask for each image so that only the desired part of each shot covers images below it. I can also fine-tune the tones of each image and feather the edges of the masks for a perfect blend at every seam.

Taking the KAP shots in the field is fast -- I shot hundreds of images covering large, complex sites in just two to three hours -- but a bit tricky. The person flying the kite cannot accurately estimate how far away the camera is. Positioning the camera requires a second person. This person may stand in the center of the desired shot and direct the kite flier to move around until the camera is directly overhead, then click the shutter half a dozen times when the camera is relatively still. Alternatively, the second person can stand outside the desired picture area, looking perpendicularly to the line of the kite. The kite flier can position the camera correctly relative to his/her right and left, and the second person can direct the kite flier to back up or walk forward until the camera is over the desired spot from his/her viewpoint. Inexpensive two-way radios help with this procedure. Sometimes cliffs, rivers, tethered bulls, and other obstacles may prevent the kite flier from standing in the spot necessary to get the camera where it is needed. By reeling in or letting out line, it is sometimes possible to work around such limitations.

The bottom line is that KAP provided extremely useful aerial views of not only the sites my project will map and excavate this summer, but also of several others that I visited. KAP has many other applications, and is a fun way to get interesting images.

Web links:

KAP E-Resources, an excellent site for KAP information in general, with many links:
www.kaper.us

An outstanding personal KAP site by an architecture professor at UC Berkeley:
www.arch.ced.berkeley.edu/kap/kaptoc.html

A good source for all kinds of kites, including the Sutton Flowform 16 that I use, as well as a generic KAP mount kit (note that the radio controlled airplane parts are not included):
www.intothewind.com

Hugin (including Panotools) public domain image correction and stitching software:
hugin.sourceforge.net

Bruce Owen (Sonoma State University) has been working in Andean archaeology since 1983, finding, mapping, and excavating sites from the central highlands to the far south coast of Peru. His website is http://www.bruceowen.com.