See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/340293905

Photogrammetry is for everyone: Structure-from-motion software user

experiences in archaeology

Article in Journal of Archaeological Science Reports · April 2020

DOI: 10.1016/j.jasrep.2020.102261

CITATIONS READS
45 434

2 authors, including:

Elizabeth Church
UCL
4 PUBLICATIONS 55 CITATIONS

SEE PROFILE

All content following this page was uploaded by Elizabeth Church on 24 July 2021.

The user has requested enhancement of the downloaded file.

Forensic Anthropology
DOI: 10.5744/fa.2020.0039

RESEARCH ARTICLE

The Forensic Utility of Photogrammetry in Surface Scene

Documentation
Elizabeth Churcha, b ● James T. Pokinesb, c ● Christopher J. H. Amesd, e

ABSTRACT: Outdoor crime scene documentation needs to be accurate and precise to preserve evidence. Photogrammetry is a poten-
tial option. Structure from Motion (SfM) processes photographs into 3D models. As commercial software does not disclose this process,
this documentation technique could be legally inappropriate. A potential solution to this problem is open-source software. A series of
mock outdoor crime scenes were documented using SfM and total station mapping. Ten large surface scatter scenes containing plastic
human remains and personal objects were laid out in 10 × 10 m units in a New England forested environment. The small surface scatter
scenes consisted of a pig (Sus scrofa) mandible placed in different environments. The resulting models were built using PhotoScan by
AgiSoft and MicMac by IGN. Accuracy was measured by the amount of variance in fixed-datum measurements, whereas visual quality
was determined by comparison.
The average total variance in fixed-datum lengths for six of the ten scenes was below 0.635 cm. The maximum differences in measure-
ment between the total station and software measurements were 0.0917 m (PhotoScan) and 0.178 m (MicMac). Comparative histograms had
low standard deviations and mean distances between points. Conditions such as light, ground foliage and topography affect model quality.
This research shows that SfM has the potential to be a rapid, accurate and low-cost resource, but there are limitations that must be
considered.

Introduction documentation. Holistic mapping strategies within forensic

archaeology can help to identify spatial patterns, increasing
the likelihood of successful human remains recovery (Moore
Overview et al. 2002). Common documentation techniques involve pho-
Evidence recovery is a destructive process. This inherent tography, sketches, and data collected with a total station (De
characteristic of recovery limits the evidentiary value of a Boer et al. 2019). Human error and subjectivity influence the
specific context once it is disturbed (Callieri et al. 2011; quality of collected data, causing degradation of information
Howland et al. 2014). Pressure is placed on an investigator to (Leeuwe 2017).
choose the appropriate recording method, in terms of time Structure-from-Motion (SfM) photogrammetry is
and budget constraints, while ensuring comprehensive another potential technique for scene documentation,
although it is not yet established in common forensic prac-
tice (Baier & Rando 2016; De Boer et al. 2019). SfM is a
modeling technique that uses 2D photographic datasets
a
Applied Archaeology, University College London, Institute of
Archaeology, London WC1H 0PY, UK
(photosets) to render a 3D reconstruction of the scene. This
b
Forensic Anthropology Program, Boston University School of method is a low-cost and low-time solution to digitization. It
Medicine, Department of Anataomy and Neurobiology, Boston, MA requires only a computer, digital camera and the relevant
02118, USA processing software (Magnani et al. 2016). The attractive-
c
Office of the Chief Medical Examiner, Boston Medical Center, ness of SfM in forensic archaeology is due to the ease and
Boston, MA 02118, USA
accessibility of the technique. Any investigator, not just a
d
Centre for Archaeological Science, School of Earth, Atmospheric and
Life Sciences, University of Wollongong, NSW 2522, Australia
visualization specialist, would have the ability to create
e
Department of Anthropology, University of Victoria, Victoria, BC photo-realistic, spatially accurate records of entire scenes
V8P 5C2, Canada (Sheppard et al. 2017; Howland et al. 2014). Photo-realism
*Correspondence to:Elizabeth Church, Applied Archaeology,
and spatial accuracy are the two principle evaluative areas
University College London, Institute of Archaeology, 31–34 Gordon for SfM, as interpretive and metric contexts (Wolf et al.
Square, Bloomsbury, London WC1H 0PY, UK 2014). The present research examines the interpretive and
E-mail: [email protected] metric potential of SfM photogrammetry for mock, outdoor,
Received 11 August 2020; Revised 23 September 2020; surface scatter crime scenes within a New England environ-
Accepted 31 October 2020 ment. A secondary objective is to demonstrate that

© 2021 University of Florida Press

2 Church et al.

open-source SfM software produces datasets within accept- the above accuracies are highly dependent on practitioner
able error margins as compared to commercial SfM soft- experience and the machine itself. A relevant example of user
ware and total station spatial mapping. error impact within this research is the outlier measurement
in the LS 8 dataset (see below). This outlier was confirmed
when the total station measurement (135 cm) was compared to
Research Design
the physical measurement (26.8 cm). The SfM software mea-
Two types of experimental scenes were created and investi- surements were comparable at 28.1 cm (PhotoScan) and 27.6
gated in the present research: large surface scatters scenes cm (MicMac). Within the entire collected dataset, this was
within a relatively consistent environment, and small sur- the only clearly unrealistic total station measurement. See
face scatter scenes within variable environments. The goal Church (2019) for a full discussion of user error. The present
of examining large surface scatter scenes (LS) is to test the research treats the data collected with a total station as the
documentation capacity of SfM within a forested environ- “ground-truth” measurement from which the SfM dataset
ment through (1) measurement comparison and (2) point- will be compared (excluding the outlier).
cloud distribution assessments. The goals of the smaller
surface scatter scenes (SS) are (1) to test commercial and
Commercial Software in Forensics
open-source SfM capacity in rendering colors and textures
within a variety of environments and (2) to examine SfM The use of SfM photogrammetry within digital archaeology
precision in generating models of the same environment is well established (Graham et al. 2020). PhotoScan by
across different photosets and temporal conditions through AgiSoft is a popular commercial software used within cul-
point-cloud distribution analysis. LS data were gathered tural heritage management (Discamps et al. 2016; Douglass
from the creation of ten surface scatter mock crime scenes, et al. 2015; Ducke et al. 2011; Georgopoulos 2016; Howland
while the SS data were gathered from the placement of a et al. 2014; Koenig et al. 2017; De Reu et al. 2014; Sapirstein
skeletonized pig (Sus scrofa) mandible within different con- & Murray 2017; Zaitceva et al. 2016). Recent validation and
texts, over two data collection trials. workflow studies of close-range SfM photogrammetry
within academic forensic archaeology (Baier & Rando 2016;
Carlton et al. 2017; Edelman & Aalders 2018; Ferrell 2020;
SfM Metric Potential
Gidusko et al. 2018; Mccollum 2020; Omari 2019) have
Measurement sensitivity is an important component of vali- cemented PhotoScan as a popular software option. Rivero
dation analysis. In order for a map to be meaningful, it must et al. (2019) refer to PhotoScan as the “de facto standard for
be accurate and realistic (Berezowski et al. 2020). The close range photogrammetry” within archaeology (4).
International Organisation for Standardisation (ISO) and “Close range” means that photosets were gathered by a pho-
the Forensic Science Regular (FSR) provide specific valida- tographer at a short distance from the scene. Long-range
tion assessment requirements to ensure a method is “fit-for SfM photogrammetry through the use of drone or satellite
purpose” (FSR 2016; 32) and has a defined limit of accuracy imagery has similar validation studies (Abate et al. 2019),
(FSR 2016; 37–39). In terms of forensic mapping, published but that is beyond the scope of the present research. The
limits of accuracy are rare. The need to establish statistical appropriateness of commercial software like PhotoScan
frameworks with known error rates, with an additional within digital forensics has become a point of discussion
focus on standard operating procedures (SOPs), is an ongo- (Dennis 2020; Hynes 2018; Ruotsala 2016). All photogram-
ing discussion within forensic archaeology and anthropol- metric workflows have the same fundamental steps. Models
ogy (De Boer et al. 2020; Holobinko 2012). One of the few are produced from a reference photoset by a software appli-
recommended practice guidelines for metric accuracy cation that employs triangulation to match specific features
comes from the Technical Working Group on Crime Scene across individual photos (Jaud et al. 2016). These features
Investigation (2013). During scene processing measure- are arranged into a point-cloud which serves as a shape for
ments should be “accurate to within ¼ inch” or 0.635 cm the mesh and texture to layer upon. Depending on internal
(TWG 2013:25–26). Following FSR and ISO (17020) valida- parameters and algorithmic choice, the final model can vary
tion assessment guidelines, the published ±0.635 cm thresh- metrically and visually. For example, a point-cloud can be
old can act as a limit of measurement accuracy when noisy, meaning that the program is unable to distinguish the
compared to the “ground-truth.” features of one datum from another due to similarities, cre-
The use of a total station for mapping with forensic ating coverage gaps within the meshed images (Budka et al.
archaeology has been a mainstream practice since at least 2016). To compensate for this, the software can smooth por-
2008 (Joice 2008; De Boer et al. 2019). For reference, when tions of the model in a best-guess process during meshing.
used effectively, total station measurements are routinely As commercial software does not disclose the adjust-
accurate to within ±0.1–0.2 cm at a close range of tens of ment algorithms, it would fail to meet the Daubert standard
meters (Arias et al. 2011; Jiménez et al. 2009; Sládek et al. and related evidentiary criteria. The precedentary history
2012) and to ±1.0 cm at longer ranges (Bissaro-Júnior et al. of photogrammetric evidence within the United States legal
2018; Georgiadis et al. 2000; Jensen and Lemée 1998; system was established in the 1960s, as a type of cartogra-
Schneider and Panich 2008). It is important to be aware that phy (United States v. Louisiana et al. 1969). Daubert
Utility of Forensic Photogrammetry (SSD) 3

challenges to the use of SfM-rendered evidence have failed 2018). The users also have the option of omitting the
on the contingency that the expert is able to explain their L1Secured function if they are satisfied with the visualiza-
choice-making (Ruotsala 2016:44). Commercial software, tion quality of the model. In a post-Daubert world, the type
like PhotoScan, does not disclose the choice-making algo- of “agile development” (Lynch & Stephen 2018:239) which
rithms in any part of model-processing (i.e., black-box soft- MicMac encourages through transparency and method-
ware) (Arza-García et al. 2019:196). The official legal ological disclosure facilitates better forensic practices.
definition of photogrammetry within casework has become
“the science of making reliable measurements by the use of
photographs” (Williams v. State 2020), which is a simplified Methods
definition. As model rendering and 3D post-production
becomes more specialized (Valente 2019), the digital
choices made become more obtuse. Forensic DNA analysis Overview
is currently facing increasing legal scrutiny due to the use Two types of outdoor mock scenes were created and
of black-box software (Matthews et al. 2019, 2020). mapped with SfM photogrammetry and a control TS. These
scenes were staged at Boston University’s Outdoor
Open-Source Software in Forensics Research Facility (ORF) in Holliston, MA. All photosets
were taken with a Nikon D5200 DSLR camera, and all
Comparatively, open-source software that discloses both the models were rendered on a consumer-grade laptop. Tennis
SfM toolchains and the computer code that manipulates the balls were placed on the scene borders to help delineate
photoset is freely available (Green et al. 2014). Multi-Images edges during SfM processing. Camera settings were placed
Correspondences: Méthods Automatiques de Corrélation on automatic in order to help negate the variable lighting
(MicMac 2008) was spearheaded as a general-purpose conditions during both in-scene and between-scene pho-
image matching tool by Cléry (2018) and developed into a tography. Photosets were aligned on-site to check for cover-
dedicated SfM matching tool in 2007. Budka et al. (2016: 22)
age gaps that can be produced if camera angles of individual
suggested five essential principles for 3D model construc-
photographs do not have enough overlap. A full discussion
tion within a forensic context:
of photogrammetric workflows, histograms, and a detailed
1. A third party should be able to reproduce all stages
breakdown of SfM rendering can be found in Church
of an analysis independently from the raw data.
(2019). PhotoScan was used to render all commercial-grade
2. A detailed file history with processing notes should
models, following Jaud et al. (2016) and recommendations
be available.
in the PhotoScan Manual (AgiSoft 2018). MicMac was used
3. In creating and processing a three-dimensional
to render all open-source models, sourced from the MicMac
model, digital files should be saved at all significant
GitHub repository, and rendering parameters following the
steps, allowing a third party to evaluate the impact
of each step. MicMac manual by Cléry (2018).
4. There should be complete transparency at all times Cloud Compare (2019) was used to generate all point-
in terms of data limitations and risks to accuracy. cloud histogram comparisons. Models were scaled with
5. All data files should be collated and made available the match-box boundary function and manually aligned.
if required including the raw data. The aligned point-clouds were registered using fine point-
These forensic principles, when combined with the Daubert cloud registration (ICP), and cloud-cloud distances com-
standard, make open-source SfM the more viable option puted using a cloud-mesh distance function. The average
over commercial software. For example, PhotoScan offers point-cloud distances and the standard deviation between
depth filtering which removes outliers in the point-cloud the model pairs were calculated as a histogram. These his-
automatically. The user has four filtering options when ren- tograms represent the active scalar fields of the two models
dering a model in PhotoScan, which are mild, moderate, being compared. The y-axis represents the number of
aggressive, or to disable the function entirely. A model pro- points (“count”), while the x-axis represents the deviation
duced with mild depth filtering (with the intention of trying of the same point from each other in the two point-clouds
to preserve small details, e.g., an individual roof tile) will (C2M signed distances) in cm. One of the models is arbi-
look significantly different than a model produced with trarily selected to be the base point-cloud to which the
aggressive depth filtering (large details are preserved, e.g., other point-cloud is then compared. The deviation in spa-
the roof itself) (AgiSoft 2018). MicMac allows the user to tial orientation from the point in the base point-cloud is
compensate for outliers in the point-cloud by adjusting displayed along a color spectrum stretching from negative
points repeatedly to find pixel reprojection error and then (blue) to positive (red). Gaussian (normal) distributions
applying a weighting function (L1Secured) (Cléry 2018). have been fitted to each histogram. Two similar models
An outlier is identified if the pixel reprojection error of that will have a cluster or peak of values at or near zero, result-
point exceeds a specific threshold value (EcartMax) (Cléry ing in a leptokurtic distribution. Less similar models will
4 Church et al.

have a wide range of values, displaying a more normal between scenes, starting with tightly packed surface scat-
distribution. ter and ending with wide surface scatter. Each scene
CloudCompare’s point-picking tool was used to mea- included a photography scale (101 cm), a partial set of
sure distances on the scaled models. LS scenes were docu- plastic human skeletal remains, and personal items that
mented using SfM and the total station, while SS scenes often are found within surface scatter (i.e., a wallet, tran-
were only documented with SfM (Church 2019). Success of sit card, jeans, and sneakers) (Figure 3; Table 1). Adhesive
model reconstruction varied between scenes and photosets. markers (2 cm in diameter) were placed on the objects as
LS 6 was removed from analysis due to overall poor render- measurement landmarks. These landmarks were placed
ing quality. For the SS scenes, out of a possible 32 total mod- following the rough shape of the object, and are refer-
els, 28 were confidently rendered. A single model of poor enced in Table 1 as “data points.” Coordinates of the mea-
quality was produced with PhotoScan for SS 4 after multiple surement landmarks were plotted in AutoCAD, and
alignments and on the lowest settings possible for the Trial 1 measurements were taken with the geometric distance
photoset. Alignment could not be achieved with MicMac, tool. AutoCAD measurements (i.e., total station measure-
and PhotoScan alignment for the Trial 1 photoset required ments) and SfM measurements were normalized. This
the placement of manual points to help with tie-point stabi- was done by calculating two ratios:
lization. Rendering of the model was not possible using the Ratio 1 = (SfM measurement/AutoCAD measurement)
Trial 2 photoset for either software type. The output mesh Ratio 2 = (AutoCAD measurement – SfM measurement)/
and texture for these were amorphous and not photorealistic AutoCAD measurement
(Figure 1). Standard deviations, confidence intervals (95%), and
margins of error were calculated for each scene included in the
analysis. A paired t-test was applied to the normalized scene
Large Surface Scatter Scenes (LS)
data (Ratio 2). Average absolute differences in the raw
The ten LS scenes were established in a grid pattern, with AutoCAD distances and SfM software distances were calcu-
each scene being 10 x 10 m (Figure 2). A total station lated for each measurement set and compared the 0.635 cm
(Leica Flexline TS06) and Spectra Nomad data logger best standard cut-off. The maximum and minimum differ-
running TDS Survey Pro version 5.2 were set up over the ences in measurements taken using PhotoScan, MicMac, and
primary datum. The same system was used to map each of the total station were identified. Point-cloud replicability
the LS scenes, with the total station and data logger oper- between PhotoScan and MicMac was assessed using
ated by one of the authors (JTP) and the prism held at CloudCompare’s cloud alignment and registration tools.
mapped objects by the primary author (EC) throughout.
Each grid varied in topographic composition and amounts
Small Surface Scatter Scenes (SS)
of understory vegetation. Primary vegetation present in
the gridded area was eastern white pine (Pinus strobus) The SS scenes consisted of a skeletonized pig (Sus scrofa)
with some deciduous species present (American beech, mandible within eight contexts. Each SS scene was docu-
Fagus grandifolia). Pokines et al. (2019) utilized the same mented with SfM in two trials (referred to as Trial 1 and
grid system, and provides a more detailed description of Trial 2). The environmental compositions of the scenes
the vegetation. The material concentration decreased were chosen to have different colors, textures, and amount

FIG. 1—SS 4 is an example of a noisy environment that SfM struggles to document accurately. PhotoScan produced a Phase 1 model (left), which is
amorphous compared to scene realism (right). No MicMac model was rendered. No Phase 2 models were rendered for either PhotoScan or MicMac.
Utility of Forensic Photogrammetry (SSD) 5

FIG. 2—Schematic of the grid used to establish the LS scenes (1–10) at ORF. Each small square is 1 × 1 m. Primary datum indicated with a star.
6 Church et al.

FIG. 3—The LS 1 surface scatter scene including photograph (top), PhotoScan render (middle), and MicMac render (bottom). The surface scatter in
this scene is similar to the other LS scenes, with the concentration of surface scatter decreasing between each.
Utility of Forensic Photogrammetry (SSD) 7

TABLE 1—List of Materials Used in LS Scenes. software pairs (e.g., Trial 1 SS 1 PhotoScan compared to
Trial 1 SS 1 MicMac) were assessed using CloudCompare’s
Item Classification Data Points cloud alignment and registration tools.
Articulated left foot Plastic skeletal 3
material
Right shoe Clothing 4 Results
Ulna Plastic skeletal 2
material
Radius Plastic skeletal 2 LS Scenes (Metric)
material
Left humerus Plastic skeletal 2
At the 0.05 significance level, the only statistically signifi-
material cant differential ratio was LS 2 (MicMac) (p = 0.02)
Right humerus Plastic skeletal 2 (Table 2). The mean absolute average difference measure-
material ment for MicMac LS models was 0.847 cm (LS 8 outlier
Scapula Plastic skeletal 3 included) and 0.568 cm (LS 8 outlier excluded). The mean
material
absolute average difference measurement for PhotoScan LS
Syringe Personal item 1
models was 0.564 cm (including the LS 8 outlier) and 0.289
Pedal phalanx (4×) Plastic skeletal 4
material cm (excluding the LS 8). The absolute average difference
Calcaneus Plastic skeletal 1 measurements from LS 2 (MicMac), LS 4 (PhotoScan) and
material LS 7 (PhotoScan and MicMac) exceed the 0.635 cm thresh-
Left shoe Clothing 4 old (Figure 4).
Cranium Plastic skeletal 3 The maximum and minimum differences in mea-
material surements for each software type compared to the total
Mandible Plastic skeletal 3 station measurements were also identified (Table 3).
material
For the LS PhotoScan models, the maximum measure-
Clavicle Plastic skeletal 2
material ment difference was in LS 1 (9.17 cm). This was a mea-
Sweatshirt Clothing 5 surement on the pants, with the total station measurement
Wallet Personal item 4 being 30.27 cm and the PhotoScan equivalent measure-
Lumbar vertebra (5×) Plastic skeletal 5 ment being 21.1 cm. The maximum measurement differ-
material ence between the total station and the MicMac
Rib (3×) Plastic skeletal 6 measurement was 17.8 cm (LS 5). Like the PhotoScan
material maximum difference, this difference was also from the
Metro card Personal item 1
pants. The total station measurement for these fixed
Plastic shotgun shell Personal item 1
points was 97.39 cm, and the MicMac measurement was
Sacrum Plastic skeletal 3
material
79.6 cm. The minimum measurement difference for the
Left innominate Plastic skeletal 3
LS PhotoScan models is 0.00 cm, and can be found in LS
material 1 and LS 7, from measurements taken on the plastic skel-
Jeans Clothing 7 etal material. The minimum measurement difference for
Right femur Plastic skeletal 2 the LS MicMac models was 0.00 cm, from LS models 5,
material 8 and 9. This indicates agreement between the total sta-
Left femur Plastic skeletal 2 tion and MicMac measurements.
material
Left tibia Plastic skeletal 2
material TABLE 2—Large Surface Scatter Scene P-Values.
Photography scale Forensic 2
documentation Grid Ratio 2 Ratio 2 Number of
Number (PhotoScan) (MicMac) Measurements

1 0.54 0.54 52
of exposure of the mandible (such as a grassy area with 2 0.49 0.02 52
large pieces of overlapping and folded chicken wire, a 3 0.61 0.61 52
wooden fence with minimal foliage and a large tire track 4 0.91 0.91 52
filled with water). Point-cloud replicability was assessed 5 0.15 0.15 52
between the trial-paired models and the software-paired 7 0.15 0.15 36
models using the same method as the LS scenes. Point- 8 0.71 0.71 39
9 0.13 0.13 43
cloud replicability between trial pairs (e.g., Trial 1 SS 1
10 0.44 0.44 46
PhotoScan compared to Trial 2 SS 1 Photo Scan) and
8 Church et al.

FIG. 4—The absolute average measurement differences between raw AutoCAD distances and each photogrammetric software distance. The best
practice standard of 0.635 cm is indicated (threshold line). The x-axis indicates the scene number and the software type, while the y-axis indicates
measurement distance in cm.

TABLE 3—Measurement Differences between SfM Models and AutoCAD (cm). The LS 8 outlier has been excluded.
Maximum Minimum
Grid Number Object Object
Difference Difference
1 9.17 Pants 0.00 Pelvis
2 4.09 Pants 0.02 Right femur
3 8.42 Pants 0.07 Wallet
4 1.48 Pants 0.07 Left shoe
PhotoScan 5 8.30 Pants 0.10 Left shoe
7 0.79 Pelvis 0.00 Articulated left foot
8 3.61 Pants 0.10 Right shoe
9 1.59 Left humerus 0.10 Right humerus
10 3.31 Pants 0.20 Scapula
Grid Number Maximum Object Minimum Object
1 9.07 Pants 0.60 Right shoe
2 6.34 Pants 0.90 Radius
3 8.12 Pants 0.02 Pelvis
4 13.1 Pants 0.01 Wallet
MicMac 5 17.8 Pants 0.00 Right shoe
7 3.77 Pants 0.01 Right shoe
8 3.29 Pants 0.00 Shirt
9 2.58 Scapula 0.00 Pants
10 3.21 Right femur 0.01 Left shoe
Utility of Forensic Photogrammetry (SSD) 9

LS Scenes (Landmark Visibility) deviation was found in LS 10 (1.97 cm) as mentioned, with
the highest mean difference between point-clouds also found
The number of measurements in analysis varied between the
in LS 10 (0.39 cm). The lowest standard deviation and the
LS scenes. All 52 measurements were used in LS scenes 1
lowest mean difference were found in LS 5 (0.00013 cm and
through 5. In LS 7 through LS 10, some objects were not ren-
-0.000019 cm respectively). Therefore, the most dissimilar
dered in sufficient detail to see the adhesive marker used as
point-clouds between PhotoScan and MicMac were rendered
fixed measurement reference points. In such cases those
for LS 10 and the most similar point-clouds were rendered for
measurements were excluded from the analysis. LS 7 had the
LS 5 (Figure 6).
lowest percentage of successful measurements (69.2%), fol-
lowed by LS 8 (75%), LS 9 (82.6%) and LS 10 (88.4%).
SS Scenes (Cloud Comparisons: Software-Pairs)
LS Scenes (Cloud Comparisons: Software-Pairs) For the SS model software pairs produced in the same trials,
the standard deviations and mean distances for all scenes
Within the point-cloud comparison analysis, comparison for
were low (Table 5). Visually, the histograms for the model
LS 10 had the highest standard deviation and the highest
software pair point-clouds tend to also fall within a normal
mean distance between the pairs within point-clouds
distribution, with the largest discrepancies seen for the SS 5
(Table 4). Comparing the LS 10 histogram to the LS 7 histo-
comparisons on both (Figure 7). These discrepancies, how-
gram (Figure 5), the spread of difference is greater within LS
ever, are still minimal. The models produced by PhotoScan
10 than within LS 7. The histograms for LS scenes 1 through
and MicMac are extremely similar based on this point-cloud
9 (despite the significant difference in physical measure-
analysis.
ments for LS 2) with low standard deviations and low mean
distances between points indicate that PhotoScan and
MicMac produced very similar models. The highest standard SS Scenes (Cloud Comparisons: Trial-Pairs)
For SS 1 and SS 3 through 8, the PhotoScan trial pairs per-
formed similarly, with low standard deviations and mean
TABLE 4—LS CloudCompare Standard Deviation and Mean distances (Table 6). The models rendered show agreement
Distances (cm) (absolute values).
in the point-cloud distributions between the trial pairs and
Grid Number Standard Deviation Mean Distance were visually realistic. SS 2 has a higher standard deviation
and less agreement in the point-cloud distribution shown in
1 0.20 0.03 the histogram (Figure 8), but the point distribution does
2 0.19 0.003
generally still fall within the normal curve. The MicMac
3 0.45 0.042
model trial-pairs between Trial 1 and Trial 2 for SS 1
4 0.36 0.064
through 6 performed well along the same criteria as the
5 0.00013 0.000019
PhotoScan models. MicMac SS 2 also displayed a higher
7 0.37 0.090
standard deviation and less agreement. Unlike PhotoScan
8 0.026 0.00039
SS 7 and SS 8, the MicMac SS 7 and SS 8 show higher lev-
9 0.22 0.0065
els of point-cloud disagreement between the trial-pair mod-
10 1.9 0.39
els (Figure 9).

FIG. 5—Histograms produced for LS 10 (left) and LS 7 (right) PhotoScan and MicMac point-cloud comparison, as an example for a leptokurtic curve.
The narrowness indicating point-cloud similarity.
10 Church et al.

FIG. 6—Histograms produced for LS 10 (left) and LS 5 (right) PhotoScan and MicMac point-cloud comparison. The LS 5 histogram presents an
extremely peaked curve, indicating great similarity, while the LS 10 range shows more variation.

TABLE 5—Software-Pair Standard Deviation and Mean

Distances (cm) (absolute values). Discussion
PhotoScan
Scene Number Standard Deviation Mean Distance
Overview
1 0.16 0.034
2 0.17 0.032 The present research found that the Sf M photogramme-
3 0.11 0.016
try as a forensic documentation technique is viable
5 0.27 0.029
under specific conditions. Model rendering success
rates varied between the two types of scenes. The LS
6 0.074 0.00099
model success rate was 90% (18/20). The SS model suc-
7 0.022 0.0055
cess rate was lower at 87.5% (28/32). These two success
8 0.071 0.012
rates are comparable to Carlton et al. (2017), whose per-
MicMac
centage of unusable models was 10%. The lower success
Scene Number Standard Deviation Mean Distance
rate of the SS models was not unexpected. The environ-
1 0.27 0.12
ment of the LS scenes was more homogenous in terms of
2 0.25 0.044
colors, textures and a lack of ref lective material. This
3 0.23 0.0090 percentage of useable models is enough to address the
5 0.30 0.074 academic objective of this research. A failure to pro-
6 0.065 0.0029 duce a useable model 12.5% of the time is too high
7 0.26 0.0032 within a forensic context.
8 0.51 0.12

FIG. 7—Histograms produced for the software pair point-clouds tended to be leptokurtic. SS 5 Phase 1 (left) and Phase 2 (right) showed the largest
discrepancies between point-clouds but still retains a cluster of values at zero.
Utility of Forensic Photogrammetry (SSD) 11

TABLE 6—Trial-Pair Standard Deviation and Mean Distances (cm) (absolute values).
Scene Number Standard Deviation Mean Distance
1 0.24 0.016
2 0.60 0.21
3 0.12 0.016
PhotoScan
5 0.34 0.10
(Trial 1:Trial 2)
6 0.38 0.039
7 0.22 0.030
8 0.24 0.061
Scene Number Standard Deviation Mean Distance
1 0.25 0.034
2 0.25 0.093
3 0.39 0.16
MicMac 5 0.18 0.016
(Trial 1:Trial 2)
6 0.34 0.085
7 0.80 0.20
8 0.84 0.39

FIG. 8—Changes in the environment between dataset collection impacted trial-pair models, such as increases or decreases in reflectivity, foliage, or
light. The histograms produced for SS 8 PhotoScan trial-pair models indicate greater sensitivity to lighting changes between the trial photosets.

FIG. 9—Further examples of how environmental changes between datasets in the trial-pair models changed rendering. Between trial days, the surface
area of the water in the tire track in SS 7 expanded, and the amount of foliage present in SS 8 decreased due to landscaping. The erratic peaks in the
histograms reflect the high sensitivity of MicMac in response to those changes.
12 Church et al.

SfM Metric Evaluation trials were SS 2, SS 7 and SS 8. This was due to environ-
mental changes such as light and the amount of foliage or
In terms of “raw” measurements, the absolute average dif- water changing between the trials on which the photosets
ference exceeded the threshold in four models (4/18). This is were captured. When examining the model-pairs that use
a high failure rate within a forensic context (22%). However, the same photosets, the software types handled the scenes
when the data are appropriately adjusted to account for dif- similarly with the exceptions being changes in lighting,
ferences between scenes, the failure rate reduces consider- reflectivity, and vegetation. The capacity of the commercial
ably (6%). The outlier in LS 8 is an important example of the and the open-source software types were tested for how
total station accuracy discussion above—the assumption of they handle repeat documentation of the same scene, and
“ground-truth” accuracy changes interpretation of the abso- the results indicate that a limitation of the SfM documenta-
lute average difference measurement when compared to the tion method originates from variable environmental condi-
0.635 cm threshold. If the “ground-truth” is accurate, both tions. This was prevalent when attempting to align
software datasets perform well when normalized, with a photosets with reflective surfaces, such as the chicken wire.
94% success rate (17/18). The maximum and minimum dif- MicMac displayed more sensitivity for the environmental
ferences in raw measurements highlight the limitations differences between photosets. Comparatively, PhotoScan
within SfM software due to specific colors, textures and responded with more sensitivity when the lighting condi-
scene complexity (Carew & Errickson 2019; Peterson et al. tions were more variable between photosets.
2015). The maximum differences for both software options
were found to be on the pants. The pants used were a worn
pair of denim jeans, presenting with a soft texture, complex Software Usability
topography and dark color. Comparatively, the minimum As mentioned above, the use of open-source software in
differences for both software options came from skeletal SfM photogrammetry should be preferred to commercial
material. The SfM software likely performed better due to options, but more discussion around real-world implementa-
the higher contrasting landmarks, and the wear-patina on tion is required. Edelman and Aalders (2018:14) suggest that
the plastic surface resulted in a simple topography with a in order to produce a high-quality forensic SfM model, all
matte surface texture. that is needed are “practical guidelines, basic awareness and
rules of thumb.” Carlton et al. (2017) have similar sugges-
SfM Interpretive Evaluation tions about model production. There are multiple resources
providing archaeological SfM workflows (Sapirstein 2016,
An issue within the SfM LS dataset are the low-quality mod- 2017; Sapirstein and Murray 2017) which can be adapted to
els produced in LS 7 through LS 10. The low resolution meant gather forensic photosets. Methodological suggestions from
that not all landmarks were visible and not all measurements this research which may be particularly useful in surface
taken. As scene complexity and “noise” increased from LS 5 scatter documentation include larger numbers of photo-
onwards, this was not unexpected. This indicates that the graphs within the photosets (100+), a wide range of field
ability of SfM algorithms, both commercially and open- (3–4 m), and that individual photographers are comfortable
source, are still limited at this time, and should not be the only adjusting camera settings to accommodate for variable
documentation technique employed. The standard deviations lighting conditions. The training required to gather pho-
between the point-clouds themselves are very low, indicating tosets is relatively minimal. Students or investigators who
that the differences come from the mesh and texture. have taken forensic scene photos or laboratory photos
Increasing the rate of usable models, and thus the number of already have the required skills.
visible landmarks, can be achieved through careful consider- PhotoScan has a short-learning curve and an intuitive
ation of photoset quality and quantity (Carlton et al. 2017). GUI (graphical user interface) (see Jones and Church [2020]
Suboptimal conditions can be negated somewhat by increas- for a user experience discussion). It is distributed under mul-
ing the number of photographs within the photoset. Without tiple licenses, with varying cost points. There are profes-
enough photographs to provide addition detail, the extrapo- sional and standard editions, which provide users with
lated mesh and texture layers can appear warped or stretched different levels of access to the functions of the program.
(Carlton et al. 2017; Micheletti et al. 2015). A secondary issue A standard edition only provides basic options for modeling:
rising from this solution is that by increasing photoset size, alignment, dense point-cloud generation, mesh and texture
computing power becomes a limitation. If large photosets are generation (AgiSoft 2018). The professional edition provides
retained in an original state, as suggested by Baier and Rando valuable options such as ground control points, coded and
(2016), then this could be solved with the eventual acquisition non-coded target detection, and supports the addition of
of more memory or a desktop for rendering models at a later scale bars (AgiSoft 2018). Without scale bars or ground con-
date. trol points, the model cannot be spatially adjusted to reflect
Complex environmental changes also had a high reality and would be useless in a forensic context. Therefore,
impact on the quality of the SS scenes. The SS scenes with the professional, and more expensive version, is the essential
higher variability and therefore lower replicability between option in forensic documentation. Comparatively, MicMac
Utility of Forensic Photogrammetry (SSD) 13

was conceptualized as a photogrammetric program with a Adoption of new technologies within forensic archae-
high degree of versatility. For an individual without a back- ology can be a viable way of increasing scope of informa-
ground in terminal-based or non-GUI software programs, it tion documented. Ensuring good practice within the
requires a considerable time investment to learn (Church application of SfM photogrammetry to forensic contexts
2019). MicMac is distributed under a French CeCILL-B means recognizing the need for open-source software adop-
license. This is an internationally recognized licensing tion, which is required to meet basic transparency and legal
option, which guarantees access to the source code of the requirements. Being aware of what applications and soft-
program with “the freedom to use it for all purposes, to mod- ware outcomes are appropriate to use and reviewing the
ify it and to redistribute all modifications” (CeCILL 2019). applied licensing parameters of software models are part of
What the CeCILL-B license guarantees is that when inte- the basic consideration for digital strategies (Ulguim 2018).
grated with a software under another licence, the integration The adoption of open-source software in forensic SfM
is also distributed under CeCILL-B, ensuring that access to should be encouraged, especially within forensic anthropol-
the source code is maintained. This would mean that ogy and archaeology (Lynch & Stephan 2018:236). Active
MicMac, and open-source applications are compatible with online and academic communities are improving access to
it, and are viable long-term programs for forensic use, as software itself, and accessibility to education about that
future iterations of MicMac are protected. software will continue to improve exponentially.

Conclusions
References
The advantages of SfM photogrammetry can be best sum-
marized as a noninvasive, lightweight, and low-cost method Abate D, Colls Sturdy C, Moyssi N, Karsili D, Faka M, Anilir A,
for crime scene documentation (Carew & Errickson 2019; Manolis S. Optimizing search strategies in mass grave loca-
tion through the combination of digital technologies. Forensic
Lech et al. 2018). However, the disadvantages at this time Science International: Synergy 2019;1:95–107.
are considerable, regardless of software choice. SfM photo- Agisoft. AgiSoft PhotoScan User Manual; Professional Edition
grammetry struggles with uniform textures, mono-colored v.1.4. Electronic document. Agisoft LLC: St. Petersburg,
surfaces, and reflective surfaces. The 3D model produced is Russia; 2018.
static—it can only represent what is surface-visible. Objects, AgiSoft PhotoScan Professional (version 1.4). http://www.agisoft.
com, 2018.
evidentiary or not, that are buried, hidden, or otherwise Arias P, González-Aguilera D, Riverio B, Caparrini N.
obscured, are not represented. There is no perfect, “real- Orthoimage-based documentation of archaeological struc-
world” scene which would have flat lighting, simple, non-re- tures: The case of a Mediaeval wall in Pontevedra, Spain.
flective, structures and the ideal ratio of textures and colors Archaeometry 2010;53(4):858–872.
for SfM documentation. Arza-García M, Gil-Docampo M, Ortiz-Sanz. A hybrid photo-
grammetry approach for archaeological sites: Block align-
There are practical strategies that can negate these
ment issues in a case study (the Roman camp of A Cidadela).
disadvantages to some effect. Proper photoset capturing Journal of Cultural Heritage 2019;38:195–203.
techniques such as using coded or noncoded targets, Baier W, Rando C. Developing the use of Structure-from-Motion
ensuring right amount of overlap between photographs, or in mass grave documentation. Forensic Science International
placing markers for manual alignment within the software 2016;261:19–25.
increase the odds of a useable model. Multiple photosets Berezowski V, Mallett X, Moffat I. Geomatic techniques in foren-
sic science: A review. Science & Justice 2020;60(2):99–107.
could be taken during the search period, site processing, Bissaro-Júnior MC, Ghilardi RP, Rosario Bueno M, Manzoli A,
and site excavation and then linked together in order to Adorni FS, Muniz FP, Guilherme E, De Souza Filho JP, Negri
reveal objects under leaf litter. This approach would be FR, Hsiou AS. The total station as a tool for recording prove-
similar to “daisy-chaining” multiple alignments or chunks nance in paleontology fieldwork: Configuration, use, advan-
together that is already possible within the software tages, and disadvantages. Palaios 2018;33:55–60.
Budka M, Bakirov R, Deng S, Falkingham P, Reynolds S,
(AgiSoft 2018). The use of chunk alignment is common Underhill S, Bennett M. DigTrace Pro/Academic User
practice for archaeological sites as a form of holistic map- Manual. https://www.digtrace.co.uk/manuals; 2016.
ping (Rivero et al. 2019; Arza-García et al. 2019). Callieri M, Dell'Unto N, Dellepiane M, Scopigno R, Soderberg B,
Increasing an emphasis on transparent post-production Larsson L. Documentation and interpretation of an archeo-
would ensure the readability of scenes documented in logical excavation: Experience with dense stereo reconstruc-
tion tools. In: Dellepaine M, Niccolucci F, Pena Serna S,
suboptimal conditions (Valente 2019). For instance, vari- Rushmeier H, Van Gool L, eds. VAST 2011: Proceedings of the
able lighting conditions could be dealt with through 12th International Conference on Virtual Reality, Archaeology
CloudCompare’s mesh editing abilities and still maintain and Cultural Heritage. Goslar, Germany: Eurographics
forensically viable workflows. MicMac is user-intensive Association; 2011:33–40.
but transparent. PhotoScan is comparatively easy to use, Carew R, Errichson D. Imaging in forensic science: Five years on.
Journal of Forensic Radiology and Imaging 2019;16:24–33.
but rendering is veiled. While commercial software is
Carlton CD, Mitchell S, Lewis P. Preliminary application of
more convenient in archaeology, this is not the same case Structure from Motion and GIS to document decomposition
in a forensic context, and the time/labor investment into and taphonomic processes. Forensic Science International
an open-source option is worthwhile. 2017;282:41–45.
14 Church et al.

CeCILL. CeCILL-B Free Software License Agreement. https:// Green S, Bevan A, Shapland M. A comparative assessment of
cecill.info/licences/Licence_CeCILL-B_V1-en.html; 2019 structure from motion methods for archaeological research.
Church E. The Forensic Utility of Photogrammetry in Surface Journal of Archaeological Science 2014;46:173–181.
Scene Documentation. [MSc thesis]. https://hdl.handle. Holobinko A. Forensic human identification in the United States
net/2144/38598. Boston University; 2019. and Canada: A review of the law, admissible techniques, and
Cléry I. MicMac, Apero, Pastis and Other Beverages in a Nutshell. the legal implications of their application in forensic cases.
https://github.com/micmacIGN/Documentation/blob/master/ Forensic Science International 2012;222:394.e1–394.e13.
DocMicMac.pdf; 2018. Howland MD, Kuester F, Levy TE. Photogrammetry in the field:
CloudCompare User Manual. Version 2.6.1. http://www.cloudcom- Documenting, recording, and presenting archaeology.
pare.org /doc/qCC/CloudCompare%20v2.6.1%20-%20 Mediterranean Archaeology and Archaeometry 2014;14(4):
User%20manual.pdf. Accessed 2019. 101–108.
De Boer HH, Blau S, Delabarde T, Hackman L. The role of foren- Hynes M. Shining a brighter light into the digital “black box”: A
sic anthropology in disaster victim identification (DVI): call for stronger sociological (re)engagement with digital
Recent developments and future prospects. Forensic Sciences technology design, development and adoption debates. Irish
Research 2019;4(4):303–315. Journal of Sociology 2018;26(1):94–126.
De Boer HH, Obertová Z, Cunha E, Adalian P, Baccino E, Jaud M, Passot S, Le Bivic R, Delacourt C, Grandjean P, Le
Fracasso T, Kranioti E, Lefévre P, Lynnerup N, Petaros A, Dantec N. Assessing the accuracy of high-resolution digital
Ross A, Steyn M, Catteneo C. Strengthening the role of foren- surface models computed by PhotoScan and MicMac in
sic anthropology in person identification: Position statement sub-optimal survey conditions. Journal of Remote Sensing
by the Board of the Forensic Anthropology Society of Europe 2016;8(465):1–18.
(FASE). Forensic Science International 2020;315:1–7. Jones CA, Church E. Photogrammetry is for everyone: Structure-
De Reu J, De Smedt P, Herremans D, Van Meirvenne M, Laloo P, from-Motion software user experiences in archaeology.
De Clercq W. On introducing an image-based 3D reconstruc- Journal of Archaeological Science: Reports 2020;30:102261.
tion method in archaeological excavation practice. Journal of Jensen KS, Lemée CPP. Total station recording of large ship struc-
Archaeological Science 2014;41:251–262. tures in connection with the excavation of eight ship-wrecks
Dennis M. Digital archaeological ethics: Successes and failures in at the B&W site in Copenhagen. In Barceló J, Briz I, Vila A,
disciplinary attention. Journal of Computer Applications in eds. New Techniques for Old Times, Computer Applications
Archaeology 2020;3(1):210–218. and Quantitative Methods in Archaeology: Proceedings of
Discamps E, Muth X, Gravina B, Lacrampe-Cuyaubère F, the 26th Conference, Barcelona. Oxford, UK: Archaeopress;
Chadelle J-P, Faivre J-P, Maureille B. Photogrammetry as a 1998:85–87.
tool for integrating archival data in archaeological fieldwork: Jiménez AR, Prieto JC, Ealo JL, Guevara J, Seco F. A computer-
Examples from the Middle Palaeolithic sites of Combe- ized system to determine the provenance of finds in archaeo-
Grenal, Le Moustier and Regourdou. Journal of logical sites using acoustic signals. Journal of Archaeological
Archaeological Science: Reports 2016;8:268–276. Science 2009;36(10):2415–2426.
Douglass M, Lin S, Chodoronek M. The application of 3D photo- Joice B. Forensic Mapping: The use of total stations and mapping
grammetry for in-field documentation of archaeological fea- software to produce scale diagrams. Journal of Forensic
tures. Advances in Archaeological Practice 2015;3(2): Identification 2008;58:1–15.
136–152. Koenig CW, Willis M, Black S. Beyond the square hole:
Ducke B, Score D, Reeves J. Multiview 3D reconstruction of the Application of structure from motion photogrammetry to
archaeological site at Weymouth from image series. Computers archaeological excavation. Advances in Archaeological
& Graphics 2011;35:375–382. Practice 2017;5(1):54–70.
Edelman GJ, Aalders MC. Photogrammetry using visible, infra- Lech K, Mularczyk K, Michonski J, Januszkiewicz K, Sitnik R.
red, hyperspectral and thermal imaging of crime scenes. Novel active-SfM solution for three-dimensional crime scene
Forensic Science International 2018;292:181–189. documentation. In Bouma H, Prabhu R, Stokes R, Yitzhaky
Ferrell M. Applications of Close-Range Photogrammetry for Y, eds. Counterterrorism, Crime Fighting, Forensics, and
Documenting Human Skeletal Remains in Obstructed Surveillance Technologies II. Proc. of SPIE Vol. 108020C;
Wooded Environments [MA thesis]. Orlando: University of 2018:1–11.
Central Florida; 2020. Leeuwe R. The hiatus in crime scene documentation: Visualization
Forensic Science Regulator (FSR). Codes of Practice and Conduct of the location of evidence. Journal of Forensic Radiology
for Forensic Science Providers and Practitioners in the and Imaging 2017;8:13–16.
Criminal Justice System. Issue 3. 2016. Lynch J, Stephan C. Computational tools in forensic anthropol-
Georgiadis C, Tsioukas V, Sechidis L, Stylianidis E, Patias P. Fast ogy: The value of open-source licensing as a standard.
and accurate documentation of archaeological sites using in Forensic Anthropology 2018;1(4):228–243.
the field photogrammetric techniques. International Archives Magnani M, Douglass M, Porter S. Closing the seams: Resolving
of Photogrammetry and Remote Sensing 20000;XXXIIII frequently encountered issues in photogrammetric modelling.
(Supplement B5):28–32. Antiquity 2016;90(354):1654–1669.
Georgopoulos A. Photogrammetric automation: Is it worth it? Matthews J, Babaeianjelodar M, Lorenz S, Matthews A, Krane D,
Mediterranean Archaeology and Archaeometry Njie M, Adams N. The right to confront your accusers:
2016;16(5):11–17. Opening the black box of forensic DNA software. Paper pre-
Gidusko K, Schultz J, Branscome M. Close-range photogramme- sented at the AAAI/ACM Conference on Artificial
try applications in outdoor forensic scene documentation. Intelligence, Ethics and Society, January 27–28, 2019;
Paper presented at the 83rd Annual Meeting of the Society for Honolulu, HI.
American Archaeology, April 11–15, 2018; Washington, DC Matthews J, Northup G, Grasso I, Lorenz S, Babaeianjelodar M,
Graham S, Gupta N, Smith J, Angourakis A, Reinhard A, Bashaw H, Mondal S, Njie M, Matthews A, Goldthwaite J.
Ellenberger K, Batist Z, Rivard J, Marwick B, Carter M, When trusted black boxes don’t agree: Incentivizing iterative
Compton B, Blades R, Wood C, Nobles G. The Open Digital improvement and accountability in critical software systems.
Archaeology Textbook; 2020. Paper presented at the AAAI/ACM Conference on Artificial
 Utility of Forensic Photogrammetry (SSD) 15

Intelligence, Ethics and Society, February 7–8, 2020; New Sapirstein P. A high-precision photogrammetric recording system
York, NY. for small artifacts. Journal of Archaeological Science 2017;
Mccollum M. Refining the Data Collection Methodology of 31:33–45.
Outdoor Forensic Scenes Involving Scattered Human Sapirstein P, Murray S. Establishing best practices for photogram-
Remains Using Close-Range Photogrammetry [MA thesis]. metric recording during archaeological fieldwork. Journal of
Orlando: University of Central Florida; 2020. Field Archaeology 2017;42(4):337–350.
MicMac (1.0) [CeCILL-B software], Institut national de l’informa- Schneider T, Panich L. Total station mapping: Practical examples
tion géographique et forestière (IGN). GitHub Repository, from Alta and Baja California. Journal of California and
https://github.com/micmacIGN; 2008. Great Basin Anthropology 2008;28(2):166–183.
Micheletti N, Chandler JH, Lane SN. Structure from motion Sheppard K, Cassella JP, Fieldhouse S. A comparative study of
(SfM) photogrammetry. Geomorphological Techniques photogrammetric methods using panoramic photography in a
2015;2(2.2):1–12. forensic context. Forensic Science International 2017;273:
Moore C, Pertell D, Leney M. Analysis of pilot-related equipment 29–38.
and archaeological strategy in the recovery of aircrew losses Sládek V, Galeta P, Sosna D. Measuring human remains in the
from the Vietnam War. Journal of Forensic Sciences field: Grid technique, total station or MicroScribe? Forensic
2002;47(6):1–5. Science International 2012;221:16–22.
Omari R. Virtual Anthropology? Reliability of Three-Dimensional Technical Working Group on Crime Scene Investigation. Crime
Photogrammetry as a Forensic Anthropology Measurement Scene Investigation: A Guide for Law Enforcement. National
and Documentation Technique [MSc thesis]. Perth: Murdoch Institute of Justice. https://www.ncjrs.gov/pdffiles1/nij/178280.
University; 2019. pdf; 2013.
Peterson E, Klein M, Stewart R. Whitepaper on Structure from Ulguim P. Models and metadata: The ethics of sharing bioarchae-
Motion (SfM) Photogrammetry: Constructing Three ological 3D models online. Archaeologies: Journal of the
Dimensional Models from Photography. U.S. Department of World Archaeological Congress 2018;14(2):189–228.
the Interior, Bureau of Reclamation, Research and United States v. Louisiana et al (Texas Boundary Case). 3941 U.S 1
Development Office. https://www.usbr.gov/research/projects/ (1969). Supreme Court 1969. https://supreme.justia.com/cases/
detail.cfm?id=3835; 2015. federal/us/394/11/. Accessed 2020.
Pokines J, Purcell B, Atkinson M, Kilroy G, Udoni M, Sanders Valente R. Digital workflow to improve osteoarchaeological docu-
Stark S, Springman S, Peace B, Green M, Reinman A, Church mentation. Digital Applications in Archaeology and Cultural
E, Cassidy S, File C, Sanchez A, Herrera M. Success rates of Heritage 2019;13(e00097):1–25.
recovering teeth and infant-sized bones dispersed among leaf Williams v. State. No. 4D19-1504. Fla: Dis. Court of Appeals, 4th
litter. Forensic Anthropology 2019;2(3):168–177. Dist. 2020. https://scholar.google.com/scholar_case?case=135
Rivero O, Ruiz-López J, Intxaurbe I, Salazar S, Garate D. On the 75269272200251931&hl=en&as_sdt=2006. Electronic docu-
limits of 3D capture: A new method to approach the photo- ment. Accessed 2020.
grammetric recording of Palaeolithic thing incised engraving Wolf P, DeWitt B, Wilkinson B. Elements of Photogrammetry
in Atzurra Cave (northern Spain). Digital Applications in with Applications in GIS. 4th ed. New York: McGraw Hill
Archaeology and Cultural Heritage 2019;14:1–10. Education; 2014.
Ruotsala A. Digital Close-Range Photogrammetr—A Modern Zaitceva O, Vavulin M, Pushkarev A, Vodyasov E. Photogrammetry:
Method to Document Forensic Mass Graves [MA thesis]. from the field recording to museum presentation (Timiryazevo
University of Helsinki; 2016. burial site, Western Siberia). Mediterranean Archaeology and
Sapirstein P. Accurate measurement with photogrammetry at large Archaeometry 2016;16(5):97–103.
sites. Journal of Archaeological Science 2016;66:137–145.

View publication stats