Accurate measurement with photogrammetry at large sites P. Sapirstein Department of Art & Art History, Center for Digital Research in the Humanities, University of NebraskaeLincoln,120 Richards Hall, P.O. Box 880114, Lincoln, NE 68588-0114, USA article info Article history: Received 21 May 2015 Received in revised form 16 November 2015 Accepted 3 January 2016 Available online xxx Keywords: 3D recording Photogrammetry Accuracy Best practice Historical architecture Digital heritage Olympia abstract Photogrammetry has become increasingly popular as a low-cost method for documenting cultural heritage and archaeological excavations. However, we have yet to establish best practices for its implementation at the site, or methods for assessing the accuracy of the resulting 3D measurements. This article presents a recent study of the Temple of Hera at Olympia, where a 25  55 m area was recorded at 1 mm resolution using photogrammetry both for survey and 3D reconstruction. Coded targets were set up throughout the site, which was then photographed in two phases. First, a site-wide survey established the locations of the network of targets. Second, sets of close-up photographs for detailed 3D recon- struction of the site were registered to the global survey via the targets. This technique developed at Olympia improves measurement accuracy by an order of magnitude compared to previous imple- mentations, with a precision of at least 1 part in 50,000, and 95% of the surfaces located accurately within 2e3 mm. © 2016 Elsevier Ltd. All rights reserved. 1. Introduction Photogrammetry is all the rage these days in archaeological fieldwork. Many projects have already implemented a complete 3D recording system based on the technology (e.g. De Reu et al. 2013; De Reu et al. 2014; Dellepiane et al. 2013; Fern andez-Hernandez et al. 2014; Olson et al. 2013; Roosevelt et al. 2015; Stal et al. 2014). Modern photogrammetric systems are largely automated, using Scale Invariant Feature Transform (SIFT), Structure from Motion (SfM), and Multi-View Stereo (MVS) to restore complex 3D geometry from sets of photographs (Luhmann, 2010; Vergauwen and Van Gool, 2006). The software can rapidly create high- resolution colour 3D models of an on-going excavation or a build- ing site at much lower cost than laser-scanning hardware. Photo- grammetry now has the potential to revolutionise how archaeologists document, study, and preserve antiquity. If we are going to adopt this technology for digital recording, we should also discuss how it is implemented at the site. How should the job be set up and processed within the software? How accurate are the estimated camera positions, points, and surfaces? Massive failures are possible when the photographs do not overlap suffi- ciently, and movement within the scene during photography has unpredictable consequences. Yet with the right photographs, the software is able to create detailed 3D models that look convincingly lifelike. The beguiling realism makes it all the more critical to examine the accuracy of the results. If we are to determine best practices for photogrammetry in archaeology, assessing accuracy is essential for comparison of different implementations. As one pa- per recently published in this journal concluded, “until structure from motion can demonstrate reliable accuracy, and this can be calculated on a case by case basis, it is unlikely to be taken seriously as a measurement tool.” (Green et al. 2014, p. 181). 2. Previous research The question of accuracy is difficult to address directly, because the extensive automation of SfM/MVS software makes its operation essentially a “black box”. One approach has been to test error of individual measurements produced by SfM, typically by compari- son to a set of reference points measured with a Total Station. Ex- amples are compiled in Table 1a. For each project, an estimate of the precision has been expressed as a proportion 1:k, where k is the size of the scene divided by the reported standard error (Fraser and Brown, 1986). This metric has no inherent scale. A hypothetical camera system with a 1:5000 precision could distinguish measurements down to 1 mm across a 5-m-long vehicle, but only to 1 m when used to measure aerial E-mail address: orientalizing@gmail.com. Contents lists available at ScienceDirect Journal of Archaeological Science journal homepage: http://www.elsevier.com/locate/jas http://dx.doi.org/10.1016/j.jas.2016.01.002 0305-4403/© 2016 Elsevier Ltd. All rights reserved. Journal of Archaeological Science 66 (2016) 137e145