Geological Survey of Denmark and Greenland Bulletin 41, 2018, 99-104 99 Photogrammetry is a classical remote sensing technique dat- ing back to the 19th century that allows geologists to make three-dimensional observations in two-dimensional images using human stereopsis. Pioneering work in the 1980s and 1990s (Dueholm 1992) combined the use of vertical (nadir- looking) aerial photographs with oblique stereo images from handheld small-frame cameras into so-called multi-model photogrammetry. This was a huge technological step for- ward that made it possible to map, in three dimensions, steep terrain that would otherwise be inaccessible or poorly resolved in conventional nadir-looking imagery. The devel- opment was fundamental to the mapping and investigation of e.g. the Nuussuaq basin (Pedersen et al. 2006). Digital photogrammetry, the all-digital version of multi-model pho- togrammetry, is nowadays an efficient and powerful geologi- cal tool that is used by the Photogeological Laboratory at the Geological Survey of Denmark and Greenland (GEUS) to address geological problems in a range of projects from 3D mapping to image-based surface reconstruction and ortho- photo production. Here we present an updated description (complementary to Dueholm 1992) of the analytical pro- cedures in the typical digital workflow used in current 3D- mapping projects at GEUS. Analytical procedures for 3D mapping at the Photogeological Laboratory of the Geological Survey of Denmark and Greenland Erik Vest Sørensen and Mads Dueholm Fig. 1. During field work new stereo imagery is normally collected from helicopters or boats, but could also be collected from smaller fixed-wing aircraft, drones or while walking. © 2018 GEUS. Geological Survey of Denmark and Greenland Bulletin 41, 99–104. Open access: www.geus.dk/bulletin http://www.geus.dk/bulletin 100100 Multi-model photogrammetry in its present form is es- sentially a technique that allows geologists to combine stereo images of different origin e.g. from satellite, aerial or hand- held cameras, and with different resolutions and viewing an- gles in their geological interpretations (Fig.1), using a digital photogrammetric workstation. Examples are plentiful with scales ranging from metres (Vosgerau et al. 2010, 2016) to kilometres (Svennevig et al. 2015; Sørensen & Guarnieri 2018, this volume) or even hundreds of kilometres (Sørensen et al. 2017). The strength of the methodology lies in the ability to combine stereo images with different viewing angles. Re- gional geological structures are e.g. typically well resolved in nadir-looking aerial or satellite images, while steep cliffs are better resolved from closer range images acquired perpen- dicularly to the slope of the outcrop. Essentially, digital pho- togrammetry allows the user to map and quantify in three dimensions whatever can be seen in the stereo images across different scales and resolutions. The digital photogrammetric workstation described here is the modern equivalent to the analytical setup previously used at GEUS (Hougaard et al. 1991) and at the Institute of Surveying and Photogrammetry, Technical University of Denmark (Dueholm 1992). The workstation consists of a Windows-based computer and a split-screen 3D monitor system (Fig. 2). Technically, the monitor system displays an image of an object in one screen and an overlapping image of the same object but from a slightly different position on a second screen. A beam splitter mirror splits the polarisa- tion direction of the two screens into separate horizontal and vertical directions. Polarising glasses used by the viewer then filter the signals so that the top screen is solely presented to the right eye and the bottom screen is solely presented to the left eye. This allows the human visual system to merge the two images, whereby a stereoscopic model is created in front of the user. The central part of the workstation is the photogramme- tric software, which is essentially a computer-controlled set of algorithms that controls the viewing of the stereoscopic model as well as the collection and manipulation of three- dimensional data within the model. This allows for seamless movement and data capture between different stereoscopic models regardless of scale, origin and viewing angle, which highly increases the speed and efficiency of geological map- ping, especially in steep and inaccessible terrains. At present, the Photogeological Laboratory uses two com- mercial photogrammetric 3D-mapping software solutions (Socet GXP from BAE Systems and 3D Stereo Blend) that are complementary to each other in terms of technical capabili- ties. In the following, we describe the typical workflow used with 3D Stereo Blend: 1) data acquisition during field work, 2) data preparation of the images, 3) data interpretation (Fig. 3). This description, however, should not be viewed as a com- plete manual to the software. 3D Stereo Blend is developed by Anchor Lab in close collaboration with GEUS’ Photogeolog- ical Laboratory. The software is optimised to 3D-mapping and structural interpretation using oblique stereo images collected with calibrated hand-held digital cameras. It is an essential part of the overall strategy of the Photogeological Laboratory to increase the efficiency and usability of digital photogrammetry from data acquisition to end-product, so that the method becomes a geological tool routinely used by geologists also without prior expert knowledge. Fig. 2. The geologist at work in the Photogeological Laboratory. The stereoscopic model is displayed on a 3D monitor system that is well suited for full-day working. The stereoscopic model can be observed simultaneously by several viewers; this makes it easy to illustrate and discuss geological observations and ideas, which is beneficial for the geological interpretation. 101 Data acquisition – field photography Field work provides the geologist with important first order observations of the bedrock. Working in remote and moun- tainous areas such as Greenland is often challenging because of inaccessible outcrops (steep cliffs) and time constraints (short field seasons). The result is spatially scattered outcrop observations that can be difficult to correlate or relate to overall regional structures. To overcome this problem, stereo images, i.e. strips of overlapping images taken from different positions and covering the geological outcrops of interest are collected on a routine basis using hand-held digital cameras, commonly deployed from helicopters (Fig. 4). The simplicity of using a hand-held digital camera makes the data acquisi- tion extremely mobile and fast. Furthermore, the quality, resolution and storage capacity of modern digital cameras have led to a huge increase in efficiency and capacity during data acquisition compared to the earlier days of 3D mapping. As an example, overview images of a small outcrop were col- lected within minutes using a helicopter, while on a more re- gional scale, a 100 km cliff section was photographed within less than one hour (Sørensen et al. 2015a). Depending on the logistic setup, the images could equally well be collected by other means (Fig. 1), as long as sufficient overlap between the images is ensured. As a rule of thumb, 60–80% overlap is needed to obtain continuous overlap and good stereoscopic measurement accuracy (Dueholm 1992). However, it is now recommended that images are taken with up to 90% overlap because this results in a more successful automatic genera- tion of common points (so-called tie points) between differ- ent images. It will also, at a later stage, make it possible to use the images for surface reconstruction using dense image matching routines (Sørensen et al. 2015b). The cameras are full-frame, digital single-lens reflex with high-quality 35 mm prime lenses that are fixed and focused at infinity. However, essentially any camera can be used as long as the camera pa- rameters (lens distortion, focal length and principal point) of the camera can be modelled. The cameras are calibrated prior to field work using a test field consisting of a steel grid with c. 100 points. We recommend that the images are acquired in the raw image format of the camera and that the location of the camera is registered with Global Navigation Satellite Sys- tems (GNSS). Depending on the requirements for absolute accuracy, different GNSS equipment can be used, from sim- ple geotagging devices to more advanced differential GNSS systems. Data preparation Setting up the images essentially consists of two steps. First, a relative model is constructed by identifying common points (tie points) between overlapping images. Secondly, this mod- el is transformed into absolute, ‘real-world’ coordinates by combining camera location data and control points. Relative orientation The relative connectivity between images in object space is established by automatic tie point measurement using ‘Struc- ture from Motion’ (SfM) image-matching algorithms (Lowe 2004; Snavely et al. 2008), with possible manual editing or addition of tie points by stereoscopic measurement within 3D Stereo Blend. The commercial software Photoscan Profes- sional (PS) from Agisoft is used for the image matching, but other software solutions can also be used. The raw image- matches (tie points) are subsequently exported to 3D Stereo Blend. 3D Stereo Blend uses the imported tie point file as a block definition and set-up file as well as for preliminary triangulation. The file is typically thinned when imported using an area-based thinning technique to give an even dis- tribution of tie points across the images. The next step is to make a preliminary so-called bundle adjustment or trian- gulation, including error detection and elimination of er- roneous tie points from the imported data. The images can subsequently be viewed in 3D with the orientation of that exported from PS. If positional data, such as e.g. GNSS cam- era positions, are included in the export from PS, the stereo- Field acqusition using calibrated digital cameras Automatic Image matching using Structure from Motion (Photoscan) Traditional photogrammetric bundle adjustment (3D Stereo Blend) Archive stereo imagery Surface reconstruction (Point cloud/DEM) (SURE & Photoscan) 3D mapping (3D Stereo Blend) Data interpretation Data preparation Data acquisition Fig. 3. Schematic flow diagram summarising the typical 3D-mapping workflow from data acquisition to photogrammetric data preparation and interpretation. 102102 scopic model will be placed in absolute coordinates within 3D Stereo Blend. Depending on the project requirements, one could move directly on to the 3D-mapping stage for rapid results or if absolute accuracy is of minor importance. However, requirements for absolute accuracy often mean that additional control data must be added. Absolute orientation The absolute orientation relates the photogrammetric mod- els in object space to absolute coordinates through a proper bundle adjustment process, whereby the stereoscopic model gets the correct scale and levelling. When solving the bundle adjustment, all provided control information is weighted ac- cording to an a priori estimated error. Different sources of control data can be used in this process, including surveyed control points, pass-points, camera GNSS-data, planar lev- elling points and distances. Surveyed ground-control points generally give the best positional accuracy. However, consid- ering the regional scale of many 3D-mapping projects, this approach is often not viable from a practical point of view. Instead, pass-points from already aerotriangulated aerial photographs can be used as control-point source, which eliminates the need for ground-control-point collection dur- ing field work. This is done by identifying common points between for example a set of vertical aerial photographs and local oblique-view images. This process is sensitive due to the different perspectives of the image data sets as well as the different resolutions, and takes some practical experience to carry out. However, workflows implemented within 3D Ste- reo Blend have significantly improved the efficiency of the identification process. GNSS data collected with the camera yield the position of the camera at the time of acquisition that is important for setting up the images. Simple geotag- ging equipment typically yields a camera position accuracy of around 5–10 m. However, more advanced differential GNSS set-ups could result in a positional accuracy at the sub-metre level, which might minimise or completely eliminate the need for surveyed ground- control points or measured pass- points. The overall time consumption in the preparation of the images has significantly decreased compared to the early days of analytical 3D mapping. This is largely a consequence of computer hardware and digital camera development, but also of software improvements including better image-matching algorithms and improved photogrammetric workflows in the 3D Stereo Blend software. Consequently, small blocks of stereo images can be prepared for geological interpretation within a day. This opens up for e.g. using digital photogram- metry as an active tool during field work. Stereoscopic image interpretation – 3D mapping Once the images are properly prepared, or if archive data are available, the geologist can commence the geological pho- tointerpretation in three dimensions. This is done by trac- ing geological features of interest in the stereoscopic mod- els. The change between neighbouring stereo models takes place automatically, so that important horizons can be traced seamlessly for kilometres. Several views can be opened simul- taneously, whereby an outcrop can be seen from different perspectives and at different resolutions, which is important for the geological interpretation. The different views can fur- Fig. 4. Stereo images collected from helicopter using a hand-held digital camera. This method makes the data acquisition very mobile and fast. The location of the camera is registered with Global Navigation Satellite Systems. Photo: Jonas Petersen. 103 thermore be linked, so that a movement in one window is also updated in the linked window. Changing the displayed stereo block is either done by selecting from a pull-down win- dow or by selecting the cameras in perspective view interac- tively. All images from the cameras within a project can be shown in the perspective view. This makes it easy to manage and move around in large regional data sets with thousands of images. The outcome of the drawing is a number of vec- torised lines in 3D with many nodes (so-called polylines) su- perimposed on the stereoscopic model (Fig. 2). The polylines can be labelled and grouped according to the user’s need, and subsequently exported to GIS packages for GIS analysis or to 3D-modelling software. The software automatically reg- isters the original stereo model in which each node of a given polyline was drawn. This makes it possible to automatically adjust the images if their orientation is changed at some point. This is helpful e.g. if an initial interpretation of the images is done on a preliminary set-up of the images without proper ground control. A central part of the 3D Stereo Blend software is a set of structural tools that enables the viewer to evaluate structural parameters such as strike and dip of bedding, plunge and di- rection of fold axes and stratigraphic thickness of beds. This allows the geologist for instance to populate a geological map with many structural observations from areas that could not be visited in the field, to correlate bedding from one side of a fjord to the other or to measure true thicknesses. The struc- tural tool-set consists of selected routines from the Geopro- gram software (Dueholm & Coe 1989) and works by fitting planes by least-squares adjustment of captured data points. Of special importance is the possibility to project captured data into geological sections which can have arbitrary orientations and also be inclined to better reflect e.g. true thicknesses. In summary, with the present set-up, 3D mapping has be- come much more effective and user-friendly than previously, which is largely due to the improved photogrammetric work- flows in the 3D Stereo Blend software. For instance, it is now possible for the untrained geologist to engage in 3D mapping with only 1–2 days of training. Resolution and accuracy The resolution of the stereoscopic models depends on the distance between the camera and outcrop, the camera focal length and the pixel size (pixel pitch) of the sensor of the digi- tal camera. As an example, photographing an outcrop at a distance of 100 m gives a ground sampling distance (GSD) of 14 mm using e.g. a 36 megapixel Nikon D800E camera with a calibrated sensor pixel pitch of 4.89 µm and a focal length of 36 mm, or a scale of c. 1:3000, while increasing the dis- tance to 1000 m will give images with GSD of 0.14 m (scale 1:30 000). The photogrammetric or geometric accuracy in the image plane on the ground (the x and y axes perpendicular to the direction of view) relates to the distance between the cam- era and outcrop, the camera focal length and how accurately a point can be determined in the stereoscopic model. The latter is a function of the accuracy of the triangulation, the camera calibration and on how well the user can place a point in the stereoscopic model. In addition, the photogrammetric accuracy in depth (the z axis, in the direction of view) also de- pends on the ratio between the distance between the camera and the object and the distance between the camera stations (also called the baseline). A typical value for the point deter- mination is around one pixel, which leads to an accuracy of c. 14 mm in the image plane, while the accuracy in depth is c. 35 mm with a distance to the outcrop of 100 m and a base- line of 40 m (corresponding to 60% overlap). If the camera- to-object distance is increased to 1000 m, the accuracy in the image plane decreases to 0.14 m, while the accuracy in depth will decrease to 0.35 m, assuming that the baseline is increased to 400 m to maintain a 60% image overlap. How- ever, if the baseline remains 40 m long there is a significant decrease in the depth accuracy to c. 3.5 m. This illustrates the importance of having an appropriate image overlap. The absolute accuracy of the stereoscopic model relates to that of the control source. In Greenland, pass-points are commonly acquired from the monochrome vertical aerial photographs on a scale of 1:150 000 that cover most ice-free areas. Typical achievable accuracies on the point transfer from these photographs is around 3–5 m. Using high-preci- sion differential GNNS setup it should be possible to obtain accuracies of less than 1 m, whereby the absolute accuracy approaches the photogrammetric accuracy. Although the absolute accuracy generally exceeds that of the photogrammetric accuracy, the relative accuracy be- tween models remains equal to the photogrammetric accu- racy, because the ground-control data is weighted during the bundle adjustment. This means that when calculating e.g. thicknesses or structural parameters such as strike and dip, the accuracy is determined from the photogrammetric accu- racy. In practice, this is all handled automatically within the 3D Stereo Blend software. Other derived products In addition to the 3D geological mapping workflow, the tech- nical development of automatic multi-view- stereo-matching routines (Rothermel et al. 2012) has facilitated the extrac- tion of digital outcrop models from stereo imagery (Sørensen 104104 et al. 2015b). The digital outcrop model is a 3D representa- tion of the outcrop surface. This type of routine utilises the redundancy of high image overlap to produce a set of high- resolution data points in space (so-called point clouds) that need little manual editing. Although a high image overlap (i.e. a small baseline) reduces the precision of individually matched pixels, this is compensated by determining the same point in multiple images, which leads to effective automatic elimination of erroneous points. The point cloud is used e.g. for visualisation purposes where it can be integrated with the results of the 3D geological mapping, but it can also be further processed into 3D mesh representations of the ter- rain, or production of digital terrain models and orthopho- tos. GEUS’ Photogeological Laboratory is currently using a suite of software solutions for terrain extraction. Software such as Photoscan Professional from Agisoft and SURE from Nframes is typically used in the digital outcrop model work- flow, while more conventional aerial and satellite imagery is processed using Socet GXP. Summary The all-digital version of multi-model photogrammetry, now referred to as digital photogrammetry or just 3D mapping, has brought the geological outcrop into GEUS’ Photogeo- logical Laboratory. The recent increases in efficiency all the way from data acquisition to the geological interpretation makes 3D mapping an attractive geological tool available to the geologist. With a digital photogrammetric workstation, the users can view, map and explore any geological feature in three dimensions, following the principle that whatever can be seen in the images can also be mapped and quantified in 3D. Acknowledgments The authors would like to thank Max Strunck and Birgir Óskarsson for re- viewing the manuscript as well as Asger Ken Pedersen and Lotte Melchior Larsen for helpful comments and suggestions. References Dueholm, K.S. & Coe, J.A. 1989: Geoprogram. Program for geological photogrammetry. The Compass 66, 59–64. Dueholm, K.S. 1992: Geologic photogrammetry using standard small- frame cameras. In: Dueholm, K.S. & Pedersen, A.K. (eds): Geological analysis and mapping using multi-model photogrammetry. Rapport Grønlands Geologiske Undersøgelse 156, 7–17. Hougaard, G., Jepsen, H.F. & Neve, J.K. 1991: GGU’s photogeological laboratory: aerial photogrammetry – a valuable geological mapping tool in Greenland. Grønlands Geologiske Undersøgelse Rapport 152, 32–35. Lowe, D.G. 2004: Distinctive image features from scale-invariant key points. International Journal of Computer Vision 60, 91–110. Pedersen, A.K., Larsen, L.M., Pedersen, G.K. & Dueholm, K.S. 2006: Five slices through the Nuussuaq Basin, West Greenland. Geological Survey of Denmark and Greenland Bulletin 10, 53–56. Rothermel M., Wenzel, K., Fritsch, D. & Haala, N. 2012: SURE: pho- togrammetric surface reconstruction from imagery. In: Proceedings, LC3D Workshop, Berlin, 4–5 December 2012, 9 pp. Snavely, N., Seitz, S. & Szeliski, R. 2008: Modeling the World from in- ternet photo collections. International Journal of Computer Vision 80, 189–210. Sørensen, E.V., Pedersen, A.K., Garcia-Sellés, D. & Strunck, M.N. 2015a: Point cloud from oblique stereo-imagery: an outcrop case study across scales and accessibility. European Journal of Remote Sensing 48, 593– 614. Sørensen, E.V., Bjerager, M. & Citterio, M. 2015b: Digital models based on images taken with handheld cameras – examples on land, from the sea and on ice. Geological Survey of Denmark and Greenland Bulletin 33, 73–76. Sørensen, E.V., Baker, N.G. & Guarnieri, P. 2017: Three years of photo- graphing – extreme 3D mapping. Abstract, GRSG 28th International Annual Conference – Applied Geological Remote Sensing. https:// www.grsg.org.uk/wp-content/uploads/2017/12/GRSG-AGM-and-Con- ference-Abstract-Book-2017.pdf Sørensen, E. V. & Guarnieri, P. 2018: Remote geological mapping using 3D photogrammetry: an example from Karrat, West Greenland. Geological Survey of Denmark and Greenland Bulletin 41, 63–66 (this volume). Svennevig, K., Guarnieri, P. & Stemmerik, L. 2015: From oblique pho- togrammetry to a 3D model – Structural modeling of Kilen, eastern North Greenland. Computers & Geosciences 83, 120–126. Vosgerau, H., Guarnieri, P, Weibel, R., Larsen, M., Dennehy, C., Søren- sen, E.V. & Knudsen, C. 2010: Study of a Palaeogene interbasaltic sedi- mentary unit in southern East Greenland: from 3-D photogeology to micropetrography. Geological Survey of Denmark and Greenland Bul- letin 20, 75–78. Vosgerau, H., Passey, S. R., Svennevig, K., Strunck, M. N. & Jolley, D.W. 2016: Reservoir architectures of interlava systems: a 3D-photogram- metrical study of Eocene cliff sections, Faroe Islands. Geological Soci- ety, London, Special Publications 436, 55–73. Authors’s addresses E.V.S., Geological Survey of Denmark and Greenland, Øster Voldgade 10, DK-1350 Copenhagen K, Denmark. E-mail: evs@geus.dk. M.D., Anchor Lab K/S, H.C. Andersens Boulevard 37, 5. MF, DK-1553 Copenhagen V, Denmark. https://www.grsg.org.uk/wp-content/uploads/2017/12/GRSG-AGM-and-Conference-Abstract-Book-2017.pdf https://www.grsg.org.uk/wp-content/uploads/2017/12/GRSG-AGM-and-Conference-Abstract-Book-2017.pdf https://www.grsg.org.uk/wp-content/uploads/2017/12/GRSG-AGM-and-Conference-Abstract-Book-2017.pdf mailto:evs@geus.dk