Page 8 - i1052-5173-31-9
P. 8
result is a transformation matrix indicating Two data sets, (i.e., photos and images discrepancy in the estimated horizontal plane
that to align the two point clouds, a scaling extracted from a video sequence) have been between the two models, considering the
factor of 1.0012 is required. The rotations tested to produce and later register the Photo orientation of the fault, is 1.1° around the
around the X, Y, and Z axes are –0.38°, and Video models, respectively. These models strike direction and 0.29° around the horizon-
1.00°, and 0.34° (1.1° around the strike direc- have been compared together and with the tal direction perpendicular to the fault’s strike.
tion and 0.29° around the horizontal direc- Reflex Model, which represents a benchmark In other words, the registration of the horizon-
tion perpendicular to the strike). build with photos obtained in 2016, although tal plane is sensitive to the orientation of the
probably minor morphological changes due to photographs, so that the inclusion of oblique
DISCUSSION weathering can have occurred since then. to the scene photographs may improve the
We have described a workflow for gener- Manual alignment of the Photo and Video “horizontalization” of ρ.
ating georeferenced 3D models of geologi- models shows that discrepancies ranging
cal outcrops ranging in size from tens of from 0 to 5 mm occur between the surface CONCLUSION
meters down to a few centimeters. The reconstructions. There are notable discrepan- This paper faces the need encountered by
required tools are extremely portable. Their cies between the Video and Reflex models, many field geologists to efficiently capture
use in the field is straightforward, with sur- whereas the Photo and Reflex models are images of outcrops with ultra-portable tools
vey acquisition taking a few minutes for our much more comparable, with surface dis- to produce detailed, scaled, and properly ori-
case study. During the development and placements ranging between 0 and 2 mm. ented “pocket” 3D digital representations of
testing of the procedure, it was notable that Despite the lower number of input photos, the rock exposures. Submillimeter point-cloud
video sequence acquisition can provide a Photo Model outperforms the Video Model in resolution is achieved with the suggested
more coherent scene, assuming that the terms of accuracy. The major reason for this is procedure, equaling that of models obtained
mapped area is relatively continuous. On the problematic reconstruction of the scene by means of reflex cameras, and proving the
the other hand, video sequences may gener- from extremely narrow baseline images efficiency of the proposed registration proce-
ate excessive scene overlap, complicating extracted from the video sequence. Despite dure for several quantitative applications in
image matching. Also, the use of video the video capture having a more straightfor- geology (e.g., fracture and fault orientation
frames implies the lack of control on shutter ward acquisition procedure, it may require a and associated kinematic indicators, bedding
speed, aperture, ISO, etc., limiting the use more complex and time-consuming user- attitude and thickness, fault roughness, etc.).
of video frames mostly to small outcrops. assisted procedure of image selection and Furthermore, the proposed method is intui-
Thus, selectively captured still images gen- repeated runs of photo alignment. tive so that it can be applied by all geoscien-
erally ensure a better result and a shorter Apart from minor differences in recon- tists irrespective of background or experi-
processing time, as long as the acquisition is struction quality and errors that may arise ence. In this regard, we hope that this
correctly carried out. Video models instead from manual detection of the key points used workflow will favor the widespread use of
provide a simpler acquisition scheme, albeit in the similarity transform, the registration 3D models from smartphones.
with greater risk of reconstruction artifacts. procedure of the two smartphone-generated
Once the models are built, post-process- models led to models with consistent orienta- REFERENCES CITED
ing registration using the proposed method tion and scaling characteristics. In detail, we Allmendinger, R.W., Siron, C.R., and Scott, C.P.,
is also straightforward for geoscientists observed a rotation about the vertical axis of 2017, Structural data collection with mobile de-
with limited knowledge of geospatial data 0.34°. This error, which mostly relates to digi- vices: Accuracy, redundancy, and best practices:
processing and analysis. From a practical tization of the reconstructed CH placed within Journal of Structural Geology, v. 102, p. 98–112,
https://doi.org/10.1016/j.jsg.2017.07.011.
point of view, the use of a low-cost, light- the scene, is negligible for many geological Bellian, J.A., Jennette, D.C., Kerans, C., Gibeaut, J.,
weight gimbal smartphone stabilizer offers applications, particularly if compared with Andrews, J., Yssldyk, B., and Larue, D., 2002,
a key improvement to similar workflows the accuracy of analog compasses (e.g., 3-dimensional digital outcrop data collection and
proposed previously (e.g., Tavani et al., Allmendinger et al., 2017). Such minimal analysis using eye-safe laser (LIDAR) technolo-
gy: American Association of Petroleum Geolo-
2020), and it is encouraged that geoscien- value, however, does not reflect field mea- gists Search and Discovery Article 40056, https://
tists who want to replicate the presented surement accuracy, since only one measure- www.searchanddiscovery.com/documents/ beg3d/
acquisition strategy include this item as ment was made of the same object present in (last accessed 27 Feb. 2021).
part of their standard equipment. Using a the two models. Models of the same geologi- Bemis, S.P., Micklethwaite, S., Turner, D., James,
gimbal offers two substantial advantages. cal object, created by different individuals at M.R., Akciz, S., Thiele, S.T., and Bangash, H.A.,
2014, Ground-based and UAV-based photogram-
First, stabilization of the smartphone dur- different times, could introduce additional metry: A multi-scale, high-resolution mapping
ing acquisition improves image quality rotational errors. A slight misalignment of the tool for structural geology and paleoseismology:
(i.e., by limiting motion blur), with the pro- registered horizon between the two models is Journal of Structural Geology, v. 69, p. 163–178,
duced 3D model rivaling an equivalent sur- reflected by the observed rotations around the https://doi.org/10.1016/j.jsg.2014.10.007.
face reconstruction produced with a higher x and y axes of –0.38° and 1.00°, respectively. Biber, K., Khan, S.D., Seers, T.D., Sarmiento, S.,
and Lakshmikantha, M.R., 2018, Quantitative
resolution dSLR mounted on a tripod. The This misalignment is attributable to the pro- characterization of a naturally fractured reser-
second but most fundamental advantage of cedure of horizontalization of ρ: as seen in the voir analog using a hybrid lidar-gigapixel imag-
using a gimbal is the stabilization of the rose diagram of Figure 3, ρ in both models is ing approach: Geosphere, v. 14, p. 710–730,
smartphone along its long axis, so that all clustered along a direction that is nearly paral- https://doi.org/10.1130/GES01449.1.
the images produced are oriented along a lel to the strike of the fault, providing a greater Bruna, P.O., Straubhaar, J., Prabhakaran, R., Bertot-
ti, G., Bisdom, K., Mariethoz, G., and Meda, M.,
horizontal plane, providing a constraint for constraint along the fault parallel direction 2019, A new methodology to train fracture net-
our georeferencing procedure. than along its perpendicular. Indeed, the work simulation using multiple-point statistics:
8 GSA Today | September 2021