An interactive kiosk for the Tribune del David

Text by Marc Levoy
Photographs and drawings credited individually
November 15, 2002

During our scan of Michelangelo's statues in the Medici Chapel, we briefly rotated our computer screen and let the tourists play with our 3D model. Although the real statue sat only a few feet away, tourists crowded around our computer, using it to examine parts of the statue that were difficult to see from the ground, playing with the (virtual) lighting, and generally having a good time. Inspired by this experiment, the Galleria dell'Accademia invited us to place an interactive kiosk next to the David, timed to coincide with a major cleaning of the statue planned for 2002-2004. Being given permission to install a computer display next to the world's most famous statue was a unique opportunity, so we took it very seriously. Here's the story of our kiosk, which has been operational since October 28, 2002.

The restoration

In September of 2002, the Galleria dell'Accademia began cleaning Michelangelo's giant statue of David. This cleaning was the first undertaken since the statue was moved into the museum in 1873. (Carved in 1504, the statue spent its first 350 years outdoors in Florence's Piazza della Signoria, a half-mile away.) In this photograph, we see a motorized gantry, specially constructed for the cleaning project, which gave restorers access to the statue's entire surface. The actual cleaning work was performed only when the museum was closed to the public.

To permit the restorer's gantry to roll uninterruptedly around the statue, a low platform was built flush with the top of the curb that surrounds the statue, which was also partially cut away. (We solved this same problem when we scanned the statue in 1999 by building a curb-hopping platform, shown in these pictures.) To prevent tourists from tripping on the platform, and to hide the gantry and cleaning supplies, the museum built a temporary wall around the lower part of the statue. A photograph of this wall is shown at left. Printed on the wall was a description of the restoration project. The last paragraph (at lower-right) describes the Digital Michelangelo Project.

The description of our project printed on the temporary wall stated that our 3D model would become a "data bank updated on the state of conservation of the David". This is an ambitious goal, and it will take years to achieve. However, the book Exploring David, written to document the scientific studies that preceded the restoration, made extensive use of our 3D model, and therefore marks a step in this direction. Here's a web page about the book. Our 3D model was also used to help design equipment for the restoration. At left is a computer rendering of our model, intersected with a rectangular solid showing regions of the statue that a restorer would not be able to reach using a particular gantry. If only we had had a 3D model like this to help us plan our scan of the statue!
By March 2004, restoration of the statue was almost finished. Despite the recent, highly publicized controversy about how to clean him, he looks great: lighter-colored, reasonably homogeneous, and with a "healthy glow" caused by subsurface scattering through the now-dirt-free surface. Compare this image to this general view of his head or this closeup of his ear, both taken before the restoration began.

The kiosk

Here is a plan view of the Tribune del David, showing the statue and the temporary wall around it. To the right, adjacent to one of the massive circular columns that supports the gallery roof, is our interactive kiosk. Its placement was designed to allow visitors to see our renderings and the real statue at the same time. One of the challenges we faced in designing this kiosk was the frequently crowded environment of the Galleria dell'Accademia, The kiosk must serve large numbers of people quickly, including tour groups. In the latter case, demos would be given by the tour leader; individuals would only be able to play with the kiosk during quiet periods. Another challenge we faced was the many languages that visitors speak. Written instructions were unlikely to be understood by more than a small fraction of the visitors.

The design we settled on, shown at left, consists of a 40-inch (diagonal) plasma display screen mounted slightly above eye level and a slanted control panel containing two trackballs and four buttons. There are no menus to navigate, and no cursor or text appears on the screen. One trackball is used to rotate the model, and one to move the lighting. The four buttons select different body parts to examine. One of the discoveries we made during our experiment three years ago is that the ability to zoom causes many people to get lost. To avoid this problem, and to simplify the display engine, we presented each body part at a fixed magnification.

Shown here is the control panel. Reflected in its plexiglas cover is the domed skylight that sits over the statue. The trackballs and buttons are arcade quality, in keeping with its expected heavy use. Mounted in the faceplate of each button is a thumbnail photograph of the corresponding body part. The panel is slanted to prevent people from placing handbags or backpacks on it. Hidden beneath the panel is a Dell PC with an Nvidia graphics accelerator. The PC runs Linux, customized to boot quickly and to restart the display program if it crashes (it hasn't crashed yet). The keyboard and mouse, which are not used in normal operation, are hidden from view.

The 3D models

The 3D model for each body part was extracted from a 2mm model of the statue (our raw data is at 0.29mm), simplified to 500K triangles using an out-of-core simplification algorithm developed by Cignoni, Rocchini, Montani, and Scopigno. Note that all of the models contain holes (look at his left hand), where our laser could not reach the statue surface. In keeping with the wishes of the museum director, we made no attempt to fill these holes with artificial geometry. Note that the bumps in his left forearm are not misalignments of our scan data; they show the location of repairs performed in the 1500's after the arm was broken during political riots in in the Piazza Signoria.

The models were then rendered as slightly specular, to enhance shape perception, and without surface texture. (Whatever texture we acquired three years ago was becoming obsolete as they cleaned the statue!) The inside surface was colored pink (look at the head above) to distinguish it from the outside surface. We also experimented with capping the mesh to simulate a solid, sawn-off block of marble, but people found this more confusing. The size of each model was chosen to optimize appearance for our given screen resolution, 1024 x 768 pixels, while maintaining smooth motion on our chosen hardware platform, a 1 GHz Pentium with an Nvidia Geforce4 Ti4600.

Since our magnification is fixed, we chose not to use QSplat, our multi-resolution point-based display engine. Instead, we used a simple hardware-accelerated polygon engine, which provided anti-aliasing, anti-aliased shadows, and animated transitions between the display of different body parts. Rendering time was reduced by partitioning the mesh into patches and performing occlusion culling of each patch based on a low-resolution geometric proxy. We spent a long time optimizing shadow calculations, making use of Nvidia vertex programs and register combiners.

We have just recently finished writing a secure version of this software, which permits unlicensed users to examine our models, but not extract the underlying data. Click here to download this viewer.

So does it work?

During visits to Italy in June 2003 and May of 2004, I got a chance to watch the kiosk in action, and it seems to be working well. Seeing David rotate on the raised plasma screen attracts the eye, bringing a constant stream of visitors to the kiosk. (Between November of 2002 and May of 2004, about 1.5 million people visited the Galleria dell'Accademia. Roughly one fourth of these people have given the kiosk a spin.) "Sessions" range from 30 seconds to a minute, longer if nobody is waiting. Our observerations in the Medici Chapel four years ago apply here as well: the kiosk attracts young people like flies, boys grab the controls from girls, men from women, etc. Speaking of boys, some people are awfully rough on the kiosk controls, sometimes punching the buttons and wildly spinning the trackballs. Fortunately, we used arcade-quality hardware, and so far it's holding up. Most people start with a random push on a trackball or a random button press, without reading the captions. In retrospect, real buttons & trackballs seem to invite being touched in a way that a menu and touch screen doesn't. Having limits on the allowable viewpoint (no zoom) and lighting (front hemisphere only) are very important. People are wild with the controls, but we don't allow them to zoom, so they don't get disoriented.

So is the kiosk useful? I think so. Like our experience four years ago, I watched visitors alternating between playing with the kiosk and looking at the real statue. The 3D model allowed them to explore Michelangelo's David in a new way. For a few minutes, the viewing of art became an active rather than a passive experience. By the way, we've also built a physical replica of the David, but turning it around in your hand doesn't give the same experience as viewing the model on our kiosk, and both are different than seeing the statue in person.

Credits

Many people worked on this kiosk. The software was written by Szymon Rusinkiewicz and Sean Anderson. Leslie Ikemoto and Sha Sha Chu customized the models of each body part, with help from Sean Anderson, and Paolo Cignoni simplified the models. The cabinetry was designed and built by Paolo Baldaccini of Opera Laboratori Fiorentini SRL, who also provided the drawing and photographs shown above. Construction of the kiosk was made possible by a generous donation from Pierluigi Zappacosta, founder of Logitech. Overall project direction was provided by Marc Levoy, with advice and logistical support from Roberto Scopigno and Gabriele Rossi. And of course, none of this would have happened without the cooperation and encouragement of Dr. Franca Falletti, directress of the Galleria dell'Accademia.

levoy@cs.stanford.edu

	In September of 2002, the Galleria dell'Accademia began cleaning Michelangelo's giant statue of David. This cleaning was the first undertaken since the statue was moved into the museum in 1873. (Carved in 1504, the statue spent its first 350 years outdoors in Florence's Piazza della Signoria, a half-mile away.) In this photograph, we see a motorized gantry, specially constructed for the cleaning project, which gave restorers access to the statue's entire surface. The actual cleaning work was performed only when the museum was closed to the public.
	To permit the restorer's gantry to roll uninterruptedly around the statue, a low platform was built flush with the top of the curb that surrounds the statue, which was also partially cut away. (We solved this same problem when we scanned the statue in 1999 by building a curb-hopping platform, shown in these pictures.) To prevent tourists from tripping on the platform, and to hide the gantry and cleaning supplies, the museum built a temporary wall around the lower part of the statue. A photograph of this wall is shown at left. Printed on the wall was a description of the restoration project. The last paragraph (at lower-right) describes the Digital Michelangelo Project.
	The description of our project printed on the temporary wall stated that our 3D model would become a "data bank updated on the state of conservation of the David". This is an ambitious goal, and it will take years to achieve. However, the book Exploring David, written to document the scientific studies that preceded the restoration, made extensive use of our 3D model, and therefore marks a step in this direction. Here's a web page about the book. Our 3D model was also used to help design equipment for the restoration. At left is a computer rendering of our model, intersected with a rectangular solid showing regions of the statue that a restorer would not be able to reach using a particular gantry. If only we had had a 3D model like this to help us plan our scan of the statue!
	By March 2004, restoration of the statue was almost finished. Despite the recent, highly publicized controversy about how to clean him, he looks great: lighter-colored, reasonably homogeneous, and with a "healthy glow" caused by subsurface scattering through the now-dirt-free surface. Compare this image to this general view of his head or this closeup of his ear, both taken before the restoration began.

	Here is a plan view of the Tribune del David, showing the statue and the temporary wall around it. To the right, adjacent to one of the massive circular columns that supports the gallery roof, is our interactive kiosk. Its placement was designed to allow visitors to see our renderings and the real statue at the same time. One of the challenges we faced in designing this kiosk was the frequently crowded environment of the Galleria dell'Accademia, The kiosk must serve large numbers of people quickly, including tour groups. In the latter case, demos would be given by the tour leader; individuals would only be able to play with the kiosk during quiet periods. Another challenge we faced was the many languages that visitors speak. Written instructions were unlikely to be understood by more than a small fraction of the visitors.
	The design we settled on, shown at left, consists of a 40-inch (diagonal) plasma display screen mounted slightly above eye level and a slanted control panel containing two trackballs and four buttons. There are no menus to navigate, and no cursor or text appears on the screen. One trackball is used to rotate the model, and one to move the lighting. The four buttons select different body parts to examine. One of the discoveries we made during our experiment three years ago is that the ability to zoom causes many people to get lost. To avoid this problem, and to simplify the display engine, we presented each body part at a fixed magnification.
	Shown here is the control panel. Reflected in its plexiglas cover is the domed skylight that sits over the statue. The trackballs and buttons are arcade quality, in keeping with its expected heavy use. Mounted in the faceplate of each button is a thumbnail photograph of the corresponding body part. The panel is slanted to prevent people from placing handbags or backpacks on it. Hidden beneath the panel is a Dell PC with an Nvidia graphics accelerator. The PC runs Linux, customized to boot quickly and to restart the display program if it crashes (it hasn't crashed yet). The keyboard and mouse, which are not used in normal operation, are hidden from view.