This is the seventh in a series of posts exploring 3D modeling in Mediterranean and European archaeology. For more on this project click here. We hope these papers will start a discussion either in the comments of the blog or on Twitter using the #3DMedArch hashtag.
Eric Poehler, Assistant Professor, Department of Classics, University of Massachusetts, Amherst.
Since the inception of the Pompeii Quadriporticus Project in 2010, my co-director Steven Ellis and I have been exploring the use of photogrammetry to document one of the largest monumental structures at Pompeii as part of a comprehensive digital approach to the archaeological and architectural study of this building. Our approach has attempted to integrate photogrammetry with other imaging methods, including laser scanning and ground penetrating RADAR, as well as more traditional fieldwork digital products, such as standard photography, layered vector drawings, Harris matrices, and database records. The role that photogrammetry played in this campaign has expanded exponentially over the past four field seasons (on Photogrammetry at Pompeii, see Poehler and Ellis 2012, 3, n. 6). What follows is a discussion of our experiences, our current process, and some of the pitfalls and benefits we’ve encountered.
Before discussing our use of photogrammetry, a brief introduction to the site and our project is in order. The Quadriporticus (Fig. 1), traditionally called the Caserma dei Gladiatori or Barracks of the Gladiators, is structurally rather simple: a rectangular courtyard surrounded on four sides by 74 Doric columns with dozens of small rooms behind the porticos. The building’s original design, its evolution, and its impact on the landscape of Pompeii are, however, quite complex. The Quadriporticus was also one of the earliest parts of Pompeii to be excavated. Excavation of the building had begun by February, 1767 and the clearing of the porticos and surrounding rooms continued through 1805. It was not until approximately 1819 that the great mound of debris was removed from the center of the courtyard. The Quadriporticus therefore has long served as part of both the tourist infrastructure and the tourist’s image of ancient Pompeii. Indeed, by 1792 a portion of the roof and second story had been restored and toilet facilities put in place. Today, no less than a third of the more than people 2.5 million people who visit Pompeii enter through the Quadriporticus each year (World Heritage Centre/ICOMOS report, 7-10 Janurary 2013, 43) and the only signage any of those visitors will find still reads “toilets”. These facts, redoubled by the constant threat of another major earthquake (on the 1980 earthquake damage, see Adam and Frizot 1983) that could level the site or an eruption of Vesuvius that might once again bury it completely, make our responsibility to record this building with as much detail and as much clarity as possible all the more urgent. The cork model of Pompeii, completed in 1879, stands as a testament to the value of pushing recording technologies beyond what is considered adequate for its time. With each day, more frescoes fade, more tesserae fall out of mosaics, and more walls collapse leaving the cork model as one of if not the only record of their former state. The ease, speed, and low-cost of photogrammetry has made it one of our most important tools to create a near-exhaustive spatial and visual recording of the Quadriporticus, a 21st century complement to a 19th century triumph of effort and foresight. What’s more, photogrammetry can also help us preserve the now 134 year old cork model of Pompeii as well (Fig. 2).
A hallmark of the PQP is our paperless approach (see Paperless Archaeology blog by John Wallrodt), for which we rely heavily on the iPad and a few key apps. At the same time that the iPad2 was released with a HD camera, we began a collaboration with AutoDesk Labs on their beta Project Photofly with the hope that that some basic photogrammetry might soon be possible in the field. Unfortunately, the iPad camera would not be suitable for archaeological photography until (at least) its third generation and Project Photofly, which developed into AutoDesk’s 123D Catch products (including an iPad app), required a strong internet connection to upload the necessary imagery and to download the results. The ease of making a fly-through animations when the model is complete (e.g., our Room 7, Room 40, and Room 40 with mesh view), however, remains a strength of 123D Catch. Because of the difficulty of access to an internet connection in the field, in 2012 the PQP turned to Agisoft’s PhotoScan Professional software. The software is powerful and, for the most part, intuitively designed for ease of use by non-specialists and students who are encountering the process and the program for the first time. Additionally, complex subjects can be shot as component parts (called a Chunk in PhotoScan), modeled individually, and then these parts can be themselves aligned as a complete and complex whole. 123D Catch cannot offer this option and suggests that one use another modeling project, such as AutoDesk’s own Maya. The power and ease of an out-of-the-box solution like PhotoScan, however, comes with compromises. These ultimately relate to cost. First, the professional edition costs $549 for a single educational license and a whopping $3499 for a business license. Second, PhotoScan requires a minimum of 8GB of RAM, twice what the standard laptop has installed. Certainly, this is not insurmountable, but one should look to see that her computer’s RAM is expandable before installing the program. For running image sets at the highest density settings, however, 8GB of RAM is simply too few and processes will, after many hours, simply come to a disappointing halt. Also, because archaeological images sets can be enormous (ours are c. 300MB per room, with more than 70 rooms), it is tempting to store the images on an external drive. If a PhotoScan file is opened and the images are not in the same location (say you forgot to plug in the drive), the link to the data is broken and alignment and geometry processes must be rerun. Thus, it is best to have a dedicated photogrammetry computer.
Field procedures and software processes
Our process naturally follows the contours of what the PhotoScan requires and much of that process will be similar for other programs and other archaeological environments. For example, the basics rules for capturing imagery are the same and both PhotoScan and 123D Catch and each have excellent tutorial documents and videos, respectively. For our purposes and because of our working environment, we have found that taking far more photos than is recommended to be valuable. Essentially, we ‘paint’ the subject with a large number of overview photos and then ‘walk in’ to detail shots, taking one or more intermediate images between these overview images and areas of particular interest or spatial complexity. Whenever possible, it is best to run the “Align Photos” in the lowest quality to ensure that all necessary imagery has been captured. It is particularly annoying when an otherwise complete model has a significant gap and although only one more image is needed, the subject is thousands of miles away. One benefit of shooting more images than perhaps are needed is that the missing part of a model might be found in the background of images of another subject. For the same reason, we prefer to add all possible images when running the alignment procedure because doing so both shows what images could not be automatically aligned (“NA” will appear after the file name in the image list) and what coverage the beyond the subject was also captured (Fig. 3). Such overlap is of particular importance for aligning individual models of rooms or areas since each is a small component of the Quadriporticus and will eventually all be combined into a full model of the entire building.
Having all of these extraneously matched points, however, is not always useful for the next step in the process: building the model’s geometry. Therefore, we use the cropping tool to delete these pixels before running the “Build Geometry” process for two reasons (Fig. 4). The first reason is simply to reduce processing time. The second reason is that deleting these pixels is far easier than deleting the faces that result from the build geometry process. Trimming these faces is difficult enough because of their irregular triangular shapes, but it becomes especially challenging because the selection tool captures everything in three dimensions within the area selected. Thus, not only the part one wants to remove is selected for deletion, but also everything behind it (Fig. 5). This is can be a useful point to remember because, despite how one crops pixels before building geometry, PhotoScan will sometimes close areas, such as doorways, windows, or absent ceilings, that are open in the real world. Deleting these faces therefore becomes a necessary task in realizing an accurate model of an archaeological subject, particularly architectural subjects. The final process to run in PhotoScan is to build the texture map for the model from the images used.
For complex subjects like the (at least) 96 rooms of the Quadriporticus, the individual models can and must be combined by the “align chunks” procedure. Depending on the imagery available and the degree to which it overlaps, this alignment process can be done automatically by the program, or, as in our case require the manual matching of points (See Olson et al., 2013, 254, 257).
Problems and Pitfalls
The primary problem of using photogrammetry in archaeological contexts, and one that is compounded in later procedures, is the difficulty of capturing the subjects with sufficient coverage, detail, and/or distinction. In many cases, it is impractical or even impossible to take a picture of every part of a subject or from all the angles the program might require. In the Quadriporticus, we encountered this problem in several ways. Sometimes we couldn’t get far enough away from a wall to get good overview coverage, at other times we couldn’t get close enough to get clear details, and some subjects were either partially or even nearly completely obscured. The design and the scale of the building itself became an issue. For example, because the columns and much of the façades’ masonry construction technique were so similar, the program had difficulty aligning these images and in the resulting model a wall would appear in the middle of a room and at an odd angle to the rest of the walls. Additionally, because of the Quadriporticus’ multiple stories, when we could not get high enough to see the tops of column capitals or close enough to a three story wall. This meant that the upper portions of the building were only captured from a distance or from below, at a sharp angle.
Light conditions are a perennial concern in archaeological photography and this is true for photogrammetry capture as well. In the example just discussed of the high angle photos, the sun was often an issue as the sky was always significantly brighter than the wall. Many walls of the Quadriporticus (save on those rarest of Campanian summer days – overcast, but not raining) never have full sun or full shade (Fig. 6). Inconsistent lighting is a problem not only for modeling a particular wall or even room, but also for integrating that wall or room into the complete model because at different times of day (and days of the year) the shadows cast by the architecture can be radically different. For this reason, as often as possible, we took our important photogrammetry photos at the same time each day, between noon and 1PM (contrast with decision by Olson 2013, 252, 259). We chose this hour because the shadows were shortest even though the sun was brightest and the contrast between light and shadow strongest. With much of the southern portico completely or partially reconstructed, we also faced the opposite problem: very dark, fully enclosed rooms lit only by a small window in the door or a fluorescent overhead light. Since the ambient light from the window was insufficient to provide any detail, we chose to use the overhead lights, which meant we could not capture most of these rooms’ ceilings for reconstruction.
Finally, in a building with as many visitors as the Quadripoticus has on a summer day, keeping the near background of our imagery consistent was a constant challenge. Keeping the distant background free of tourists or PQP team members was impossible. We found two solutions to this concern: 1. cropping (in Photoshop prior to adding image) or masking (in PhotoScan prior to aligning) the images to exclude extraneous features, which can give the program less to match on symmetrical objects, such as flutted columns, or 2. patience. The later, though onerous while in the moment, is far more efficient compared to masking or cropping dozens of images.
Benefits of Photogrammetry
The comparatively low cost and little training required of photogrammetry make it worthy of consideration for any archaeological investigation and an essential tool for recording a structure as large and complex as the Quadriporticus. In fact, all of our costs to use photogrammetry to record the entire building are still less than 10% of the lowest quote we received for an equivalent level of coverage using a laser scanner.
Beyond the practical aspects of recordation, there are several benefits that photogrammetic models offer to archaeological analysis and interpretation. The first is perhaps so simple as to be undervalued: in a textured model one can examine dozens or even scores of images of one’s object of interest simultaneously. Moreover, one can do this in the most intuitive of manners, by simply rotating the view like one would turn her head or body. To appreciate the value of interacting with one’s data in such a manner of organization, compare it to browsing the same image set in a folder, viewing each those images individually, and opening multiple windows (each reducing screen ‘real-estate’) to see adjacent scenes. Of far greater importance is the ability of photogrammetic models to achieve angles of viewing impossible in reality. In these virtual environments, one can – in minutes – effectively float high above or stand inside solid rock. In the Quadriporticus, the best example comes from Room 40, the only place where the extrados of a large sewer that served a large part of Pompeii (Poehler 2012, 110-11; Poehler and Ellis 2012, 9-10), still exists within the Quadriporticus. The digital reconstruction allows us to see through walls, examine the sewer’s profile, consider how much of it supports the western terrace wall, and compare that profile to other known sections of the same sewer.
At the moment of writing, laser scanning has made the total station practically obsolete for capturing data sufficient to model complex archaeological architectures. What once took the Anglo-American Project in Pompeii and the Pompeii Forum Project several years to make only 3D wire frame models, as revolutionary as they were, can now be done in a matter of weeks. Photogrammetry is also on the cusp of eclipsing the laser scanner, especially for many common recording tasks. I do not expect to see another major laser scanning campaign in Pompeii like those experiments of the last decade (Balzani et al., 2004; Cyark; Hanghai, Hori, and Ajioka 299; Hori et al., 2007). Photogrammetry, in my (admittedly lagging) opinion, is the first step in a revolution for archaeological recordation in which new techniques will offer new levels of efficiency as well as a visual and dimensional comprehensiveness to permit new forms of analysis and interpretation. Calculation of volumes – of soil excavated for normalizing ceramic assemblages or sections of masonry for estimating the materials used in their creation (e.g., Delaine 1997) – being only one of the more obvious advances. The future will continue to experiment with both large- and micro-scale uses for recording, modeling, and presenting archaeological materials through photogrammetry. For example, features recovered during excavation can be extracted from individual descriptive models (i.e., those intended to replicate drawing of stratigraphic units) and recombined with related features to better illustrate and therefore test phasing hypotheses. Such work is already being done at Gabii as reported in this blog series. Additionally, we have attempted (and thus far failed) to capture and model graffiti from walls at Pompeii, though others have succeeded on ceramics (Montani et al. 2012). The ease of creating textured, 3D images of archaeological material make the use of photogrammetry as a presentation tool more of its most potent expressions. Thus, one might photograph objects now in museums to reconstruct complete funerary assemblages and recombine them with a model of the tomb itself. Similarly, one might also use image masks and markers on pictures of frescos and mosaics now housed in museums to model the decorative as well as the spatial environment of a place by attaching those frescos and mosaics to photogrammetric models of the bare but extant masonry walls. Such reconstructive efforts are already being done for classical sculpture.
I hope that this report of our work is a modest contribution to the conversations about the place, the best practices, and the future of photogrammetry in archaeological environments.
Adam, J.-P. and M. Frizot 1983. Dégradation et restoration de l’architecture pompéienne (Paris)
Balzani M., N. Santouoli, A. Grieco, and N. Zaltron 2005. “Laser Scanner 3D Survey in Archaeological Field: the Forum of Pompeii,” International Conference on Remote Sensing Archaeology Beijing, October 18-21, 2004, 169–175. http://www.pompeiana.org/research/22-Balzani_Santopuoli.pdf
Cyark Website: http://archive.cyark.org/pompeii-intro
Delaine, J. 1997. The baths of Caracalla : a study in the design, construction, and economics of large-scale building projects in imperial Rome. Journal of Roman Archaeology, Supplement 25
Hanghai, A., Y. Hori, and O. Ajioka 2009. “Laser Scanning of Streets in Pompeii,” Proceedings of the 22nd CIPA Symposium, October 11-15, 2009, Kyoto, Japan. http://cipa.icomos.org/fileadmin/template/doc/KYOTO/166.pdf
Hori, Y., O. Ajioka, and A. Hangai 2007. “Laser Scanning in Pompeian City wall. A comparative study of accuracy of the drawings from the 1930s to 1940s,” Proceedings of 3D-Arch 2007: 3D Virtual Reconstruction and Visualization of Complex Architectures. http://www.isprs.org/proceedings/XXXVI/5-W47
Montani, I., E. Sapin, R. Sylvestre, and R. Marquis 2012. “Analysis of Roman Pottery Graffiti by High Resolution Capture and 3D Laser Profilometry,” Journal of Archaeological Science 39, 3349 – 3353.
Olson, B., R. Placchetti, J. Quartermaine, and A. Killebrew 2013. “The Tel Akko Total Archaeology Project (Akko, Israel): Assessing the suitability of multi-scale 3D field recording in archaeology,” Journal of Field Archaeology 38.3, 244 – 262.
Poehler, E. 2012. “The Drainage System at Pompeii: Mechanisms, Operation, and Design,” Journal of Roman Archaeology 25, 95 – 120.
Poehler, E. and S. Ellis 2012. “The 2011 Season of the Pompeii Quadriporticus Project: The Southwestern, Southern, Southeastern and Northern Sides.,”Fasti On Line Documents& Research 249, 1 – 12. http://www.fastionline.org/docs/FOLDER-it-2012-249.pdf
World Heritage Centre/ICOMOS report, 7-10 Janurary 2013.
Wallrodt, John. Paperless Archaeology Blog. http://paperlessarchaeology.com/