Spatial Accuracy in Orthophoto produced using UAV Photographic Images

(1)

Spatial Accuracy in Orthophoto

produced using UAV Photographic Images

Lägesnoggrannhet i ortofoton framställda med UAV-foton

Lily Ng Stensson

Faculty: Faculty of Social and Life Sciences, Geo-Science

Subject: Degree Project, Programme in Surveying and Cartography Points: Bachelor's degree 7.5 ECTS credits

Supervisor: Jan-Olov Andersson Examiner: Jan-Olov Andersson Date: 14092016

Serial number: 2016:9

(2)

Page | 2

Abstract

The popularity of using UAV in image-taking for the production of 3D models and orthophotos has increased over time. Karlskoga Municipality has recently acquired an UAV to produce their own 3D models and orthophotos. This project paper aims to study the geospatial accuracy of the orthophotos and DEM files produced using the images taken with their UAV. The flight takes only a few minutes but a considerable time is spent in the production processes. Difficulty is experienced in determining the right center point for most GCPs. Produced orthophotos in the software Photoscan have a resolution from 1.7 to 2.4 centimeters while DEM files have a resolution from 3.4 to 4.8 centimeters. Four orthophotos and four DEM files are produced where GCPs are used and not used and at two different flight heights, 76 and 105 meters.

The spatial data of the ten GCPs are identified on the orthophotos and DEM files in ArcMap and compared with GNSS NRTK measurements and Lantmäteriet's data. A visual control in terms of completeness of data, alignment, residual tilt and scale is also done. Standard deviations in plane for orthophotos there GCPs are not used are greater than 2 meters, while there GCPs are used are around 0.7 meters. Standard deviations for DEM files are observed at 0.8 meters.

(3)

Page | 3

List of abbreviations

DEM Digital Elevation Model

EPSG:5848 European Petroleum Survey Group Geodetic Parameter Dataset 5848

GCP Ground Control Point

GNSS Global Navigation Satellite System

GSD Ground Sample Distance

HMK Handbook of Surveying and Mapping Issues (Handbok i mät- och kartfrågor)

NRTK Network Real Time Kinematic (serviced by Lantmäteriet's SWEPOS satellite positioning)

RH2000 Swedish National Height System 2000 SWEREF 99 15 00 Swedish Reference Frame 1999 Zon 15^o 00^"

UAV Unmanned Aerial Vehicle

(6)

Page | 6

1. Introduction

1.1 Background

The City Planning Administration within the Karlskoga Municipality has plans in using aerial photography from UAV to create high resolution orthophotos and 3D models over an increasing number of areas in the municipality. These will serve as tools in mapping, city planning, marketing real estates and attract investments to the municipality.

Through this initiative, residents, organizations and politicians will have access to this data. They will get a better perspective and understanding over the different development plans in the municipality through the visualization of the effects of the proposed new landscapes. Likewise, the creation of orthophotos and 3D models will aid the landscape and city architects and other division officials in the development of new plans, its presentation and create a better platform in decision making for all levels. This action is also in line with the continuous development of the infrastructure for geodata in the country.

The Karlskoga Municipality provides the student with UAV photo images and GNSS NRTK measured GCPs for processing. After processing and analysis, the student will provide Karlskoga Municipality the resulting orthophoto together with a 3D model and a results analysis.

1.2 Client

The client for the project was Magnus Jordan, surveying engineer, together with Johan Mood, city planning architect of land-use and planning division (mark- och planeringsavdelningen), both working at the City Planning Administration within the Karlskoga Municipality (Samhälls- byggnadsförvaltningen, Karlskoga kommun).

1.3 Objective

The objective of the project was to produce orthophotos together with their corresponding DEMs and a 3D model over an area and evaluate both the orthophotos' and DEMs' geospatial accuracy.

1.4 Project boundary

The area of interest in this project is Äspenäs (or Espenäs), Karlskoga (60,000 m²or 6 hectare). The projected reference system concerned is in SWEREF 99 15 00, RH 2000 (EPSG:5848). Figure 1 shows the location of the study area.

(7)

Page | 7

Table 1. Spatial Bounding Coordinates in meters (SWEREF 99 15 00).

Easting (E) Northing (N) 121000 6575300 121050 6575080 120900 6575020 120780 6575150

Figure 1. Location of the study area marked in color yellow (Lantmäteriet, 2016, https://atlas.slu.se/get/).

1.5 Project Limitation

Aerial photography captured using UAV is not included in the standards and prerequisites set in the HMK-bilddata (2015) or in the HMK-ortofoto (2015). Nonetheless this project tries to follow the procedures as close to the set standards as which are applicable. The guidelines and standards set in

(8)

Page | 8

HMK Laserdata (2015) and HMK Fordonsburen laserdatainsamling (2015) will also be taken into consideration.

Selection, design and outline of the ground control points (GCPs) are done under the management of the client. Likewise, the client does all the planning and implementation of the photographing by UAV. Suggestions are presented while the client chooses the final alternative for the project. Data in form of text file and image files in jpeg format are then provided by the client for processing.

Technical documentation for GNSS NRTK measurements of GCPs is not available. A presumption in this project is that GNSS NRTK measured GCPs behold absolute accuracy according to HMK standard.

This project of creating the orthophotos, DEMs and 3D model is done with Agisoft Photoscan Pro version 1.2.4. Processing of 100-200 images requires high capacity 32-64 GB RAM, high-end graphic card and a high speed multi core CPU. It takes a considerable time to process 100-200 images and can cause program to stop functioning and thereby disrupting workflow and production efficiency.

The limitation of hardware capacity can affect and set limits to the resulting orthophotos and 3D model production.

2. Review of literature

There are several contributing factors which affect the spatial accuracy in the production of an orthophoto as taken from UAV produced images. Some factors are described in this chapter.

2.1 Ground Control Points

The ground control points or GCPs are points of known geographic location in the surveying area (Pix4d support site, 2016c). It must be easy to identify and measure. The size depends on the resolution of the image, as too small GCP can cause difficulty in identification after photo taking while too big can cause the center point being not able to be concretely indicated. According to HMK-standard 2, GCP is a square signal with 2xGSD by 2xGSD (Lantmäteriet, 2015a).

Some suggestions for the form and shape of the GCP are given in the HMK Bilddata (Lantmäteriet, 2015a) as shown in Figure 2.

Figure 2. Image samples of GCP (Lantmäteriet, 2015a).

While another company like Pix4D suggests the format shown in Figure 3.

(9)

Page | 9

Figure 3. GCP photogrammetric target (Pix4d support site, 2016c).

To get the ground sample distance (GSD), this excel computation table (Figure 4) is downloaded from Pix4D website.

Figure 4. Computation of GSD (Pix4D support site, 2016e).

where

GSD= (Sw * H * 100) / (Fr * imW) Dw= ImW*GSD/100

DH= imH*GSD/100

(Pix4d support site, 2016a, 2016c).

(10)

Page | 10

Figure 5. Sample GCP where P is pixel size (Graham and Koh, 2002).

Another source Graham & Koh (2002) has the above suggestion (Figure 5) in determining the size of a GCP.

Without GCPs, one can still produce orthophoto and 3D model but the result will have no scale, lack orientation, and no absolute position information. Furthermore, the surveyed area in 3D model may have a result of bad relative reconstruction, or simply said, it may not be able to preserve the original shape of the surveyed area. Availability of GCPs in image processing will increase the absolute accuracy of the image in its geographical location with as little variation in scaling as possible. The difference of with or without GCP is from meters to centimeters (Pix4d support site, 2016d).

With absence of constructed GCPs, natural GCP or artificial points can also be chosen like the corners of a woodlot, road intersections, or rock outcrops (Verbyla, 1995). HMK Laserdata (Lantmäteriet, 2015c) identifies corner points of objects like roads can also be used as markers just as the ridge on the roof.

In Gunnarsson & Persson's (2013) thesis paper, they were able to test the accuracy of object geographic position given different numbers and positioning of GCP. The result of their study indicates that the greater number of GCPs, the accuracy gets better. However the results between 9 GCPs and 17 GCPs differ with only 5 mm.

Agisoft (2016a) recommends at least 10 GCPs spread across the area of reconstruction.

2.2 Overlap between images

Guideline according to the HMK standard 1 and 2 indicates image overlapping at 60 % frontal and 30 % side overlapping (Lantmäteriet, 2015a). Image processing company like Pix4D (2016b) recommends 75 % frontal and 60 % side overlap in general cases while Agisoft (2016a) recommends 80 % forward overlap and 60 % side overlap.

(11)

Page | 11

2.3 Atmospheric condition - Sun position, shadowing and period of the year

The sun elevation and shadows of the objects on the images contribute to the quality of the orthophoto. Recommendation in the HMK Bilddata (Lantmäteriet, 2015a) states that sun elevation (angle/height of the sun over horizon) should be at least 30 degrees with an approximate shadow length of 1:1.7 ratio. On SMHI's homepage (2015), a sun elevation diagram is created using the analemma. Figure 6 shows the diagram for Stockholm and is applicable likewise to Karlskoga.

Photographing is recommended before birth tree leafing and after snow melting which is around May for Stockholm area. Period for photographing however varies depending on the usage of the resulting orthophoto and 3D model (Lantmäteriet, 2015a).

Figure 6. Sun elevation diagram for Stockholm area (SMHI, 2015).

2.4 Definition of an orthophoto

An orthophoto is a photographic image constructed from vertical or near vertical aerial photographs from which the distortions due to terrain relief displacement and camera tilt on the aircraft are removed (Falkner and Morgan, 2002). "An orthophoto has the same scale throughout and can be used as a map" (Orthophoto, 2016). Simply said, an orthophoto is a photographic map (USGS, 2013).

(12)

Page | 12

Positional accuracy throughout an image can attain a predictable constant quality if properly generated (Falkner and Morgan, 2002). Figure 7 shows how the effect of relief is corrected for orthophotos.

Figure 7. The effects of relief and how it is corrected for orthophotos (Falkner and Morgan, 2002).

(13)

Page | 13

3 Methodology

3.1 General Workflow in Agisoft Photoscan

Agisoft Photoscan is a computer vision software package developed by Russian manufacturer Agisoft LLC. This program uses a series of overlapping photographs to reconstruct a sparse point cloud of scenes in three dimensional shape with help of mathematical techniques such as structure from motion (SFM) algorithms together with stereo-matching algorithms (Verhoeven, 2011). The detailed processing steps used for this project can be found in Appendix 4 following the general workflow in Figure 8.

Figure 8. General workflow in Agisoft Photoscan.

3.2 Flight coverage area

The two flight sessions covered the study area as well as a bit of the surrounding area by more than 15 %.

Add photo

Align photo Build dense

cloud

Build mesh Build texture

Build DEM Build

Orthomosaic

Export DEM/

orthomosaic/

points/ 3D model

(14)

Page | 14

Figure 9. Study area in yellow dots and flight coverage area in blue line, Äspenäs, Karlskoga.

3.3 Signal form

Signal form used in this project has an A3 format box with a height of 7.8 cm as shown in Figure 10.

These markers serve as both horizontal and vertical control markers.

Figure 10. Signal form used in this project.

3.4 Images from UAV

Two flight events with different flying altitude are tested. Below are the specifications applied during each flight.

(15)

Page | 15

Table 2. Specifications used in two different flight heights, 76 meters and 105 meters.

UAV DJI Inspire 1 Professional X5 DJI Inspire 1 Professional X5

Camera Zenmuse X5 Zenmuse X5

Lens DJI MFT 15 mm DJI MFT 15 mm

Surveying application Drone Deploy Drone Deploy

Aspect ratio 4:3 4:3

Date and time June 1, 2016, noon June 1, 2016, noon

Flyght height 76 meters 105 meters

Total number of images 197 112

Duration 16 minutes 6 minutes

Speed 12 m/s 12 m/s

Area 12 hectares 12 hectares

Resolution (GSD) 1.9 cm/pix 2.6 cm/pix

Overlaps 60 % sidelap, 70 % frontlap 60 % sidelap, 70 % frontlap Reference System WGS84 (EPSG:4326) WGS84 (EPSG:4326) No. of transects/ parallel

lines

7 5

No. of GCP 8 signal form, 2 natural GCP 8 signal form, 2 natural GCP

Image file format Jpg Jpg

3.5 Placement of the ground markers

Ten GCPs are spread out over the study area at intervals as regular as possible at site as shown on Figure 11. The coordinates of the GCPs are found in Appendix 2.

Figure 11. Placement of GCPs in Äspenäs, Karlskoga. Within the yellow contour is the target area.

(16)

Page | 16

3.6 Determination of spatial accuracy

In this project, the accuracy of the position of GCPs as identified on the orthophotos which are produced with Photoscan will be compared to the existing municipal data and maps. GCPs' coordinates on ground are measured with GNSS NRTK and supplied in reference system SWEREF99 15 00 and RH2000. GCPs' plane coordinates on the produced orthophotos are identified using GIS application ArcMap. Altitude coordinates are identified in ArcMap from the produced DEM files in Photoscan and from the raster folder GSD-Höjddata, grid 2+ provided by Lantmäteriet which are also identified in ArcMap. These coordinates are tested for their absolute and relative accuracy.

3.6.1 Position Accuracy with GCP

The computation of position accuracy follows the formulas listed below:

N = northing E = easting

H = height/altitude Average deviation

Radial offset in local plane

Radial offset (per GCP)

RMS error

(Lantmäteriet, 2015b, 2015c)

Standard deviation is based on the formula:

(17)

Page | 17

3.6.2 Visual Control

The resulting orthophotos are subjected to control for completeness of content (holes), test of alignment, residual tilt of high objects and objects' boundaries.

4 Results

4.1 Orthophoto and DEM

Four orthophotos in TIFF format and four DEMs in geotiff format are produced in Photoscan and studied in ArcGIS ArcMap for their spatial accuracy. A 3D model in 3DS format is exported in Photoscan. LAS data is also produced in Photoscan and then reprocessed in ArcMap to produce a surface contour file in CAD format. One orthophoto is reproduced in FME workbench 2016 to ecw format.

The resolution for orthophotos should not be less than the specification during photographing (Lantmäteriet, 2015b) and therefore will be reduced to 0.019 m for orthophotos produced at 76 meters height and to 0.026 m for orthophotos produced at 105 meters height respectively.

Table 3. Exported file specifications for orthophotos, DEM and 3D model in reference system SWEREF 99 15 00 RH2000 EPSG:5848.

Production no.

Orthophoto File Format

Metadata Size

Resolution Pixel size

Total size in pixels 1 ExportOrthomosaic76GCP.tif

ExportOrthomosaic76GCP.tfw

1712839 kb

1 kb 0.017 m 28154x25434

2 ExportOrthomosaic76NoGCP.tif ExportOrthomosaic76NoGCP.tfw

2009661 kb

9kb 0.017 m 29385x26246

3 ExportOrthomosaic105GCP.tif ExportOrthomosaic105GCP.tfw

804438kb

1kb 0.024 m 19564x18873

4 ExportOrthomosaic105NoGCPGeneric.tif ExportOrthomosaic105NoGCPGeneric.tfw

1266541 kb

1kb 0.024 m 24350x21630

5 ExportDEM76GCPGeotiff.tif ExportDEM76GCPGeotiff.tfw

432715 kb

1 kb 0.034 m 14670x14214

6 ExportDEM76NoGCP.tif ExportDEM76NoGCP.tfw

591093 kb

1 kb 0.034 m 21625x20605

7 ExportDEM105GCP.tif ExportDEM105GCP.tfw

2102234 kb

1 kb 0.047 m 9783x9441

8 ExportDEM105NoGCPGeneric.tif ExportDEM105NoGCPGeneric.tfw

378500 kb

1 kb 0.048 m 16736x15581

9 Ecw76GCP.ecw 207166 kb

10

3DOrthoMeshTextureAdaptiveOrtho76GCP.3ds 3DOrthoMeshTextureAdaptiveOrtho76GCP.mtl 3DOrthoMeshTextureAdaptiveOrtho76GCP.jpg

1568885 kb 1 kb 3937 kb

11 LASSurfacecontour76GCP.dwg 1672 kb

(18)

Page | 18

4.2 Spatial accuracy test (Absolute accuracy)

GNSS NRTK measured GCPs are compared with the approximate GCPs identified on the

orthophotos and DEMs and have the results as shown in tables 4 and 5. The complete and detailed tables are found in appendix 5.

Table 4. Resulting accuracy test of orthophotos produced without the use of GCPs.

Table 5. Resulting accuracy test of orthophotos produced with use of 10 GCPs.

(19)

Page | 19

4.3 Spatial accuracy test (Relative accuracy)

Relative accuracy is a test in scaling by comparing the distances between two points as measured with GNSS NRTK against the corresponding points in the produced orthophotos.

Table 6. Summary of difference in distance between two GCPs in the four orthophotos in comparison against GNSS measured GCPs (in meters).

4.4 Parameters for orthophotos

Below is the table of parameters in accordance with HMK-standard 2 which is intended as guide in surveying and mapping of urban cities in the municipalities' detailed development plans and documentation (Lantmäteriet, 2015d).

Table 7. Parameters set for HMK-standard 2 as compared with resulting orthophotos produced with GCP.

Parameters HMK- standard 2 (in meters) Result from the project with GCPs Geometric resolution

(planimetric/ altitude)

0.08-0.12 / 0.10 0.019-0.026

Standard deviation (planimetric/ altitude)

0.08-0.12/0.12-0.18 0.694-0.720/0.777-0.783

Flight overlaps between transects (%) 60/30 60 % sidelap, 70 % frontlap

Color model PAN, RGB, CIR RGB

Pixel depth 8-16 bits 8 bits

Period in photographing Free from snow and leaf Free from snow

Angle of the sun (grader) <=30 Around 40 (12 noon)

4.5 GNSS measured GCP and Lantmäteriet Höjddata 2+

7.8 cm is deleted from GNSS GCP due to the box-typed signal GCP which has a height of 7.8 cm.

GCP 9 and 10 are to be ignored in this comparison for reason that Lantmäteriet Höjddata grid 2+ is based on ground level while the GNSS measured GCP 9 and 10 are not based at ground level. It must be noted that the resolution from Lantmäteriet is 2 meters per pixel. The comparison shows the deviation in GNSS NRTK measured method from Lantmäteriet's database and indicates

(20)

Page | 20

possible discrepancies as a result of the effect of resolution difference in the analysis of accuracy in DEM data.

Table 8. Comparison between GNSS NRTK measured height and Lantmäteriet Höjddata 2+ (in meters).

4.6 Visual Control

Visual control is done on the orthophotos produced with flight heights of 76 meters with GCPs and 105 meters with GCPs.

4.6.1 Test for completeness of data

The resulting orthophoto shows a number of holes throughout the image (Figure 12).

Figure 12. Example of a hole.

4.6.2 Test of alignment

The results are observed for the alignment of hanging electric power lines. The power lines are shifted from 0.5 to 1.5 meters. One type of alignment shift and distortion observed is due to vegetation (Figure 13).

(21)

Page | 21

Figure 13. Alignment error due to vegetation.

Another alignment shift is observed displacement as power lines have an offset of 1.2 meter sideways (Figure 14) on one area of the orthophoto.

Figure 14. Broken power line with offset sideways.

(22)

Page | 22

4.6.3 Residual tilt

Angular error (tilt) of electric poles is noticed (Figure 15). The height data for electric poles is not available for computation of residual tilt in this project.

Figure 15. Angular error of electric pole.

On the positive side, noticeable tilts of the houses are not observed.

4.6.4 Scaling

The resulting orthophoto does not reflect the real house boundaries from the municipality's base map (Figure 16).

Figure 16. Comparison of house perimeter from municipality's base map against produced orthophoto.

(23)

Page | 23

5. Discussion

The results in the spatial accuracy tests of the orthophotos give an error with more than the average RMS and standard deviation from past studies done by Samani (2013), Gunnarsson and Persson (2013), Axelsson and Skoog (2013) and Jansson (2013). The results neither reflect the accuracy test done by Agisoft Photoscan (Barry and Coakley, 2013) nor by Pix4D (Draeyer and Strecha, 2014).

During the assignment of ground markers in Photoscan, difficulty is experienced in finding the center point for the GCP (see Figure 17 and Figure 18). It is impossible to identify the correct center point of the GCPs as most appear to be blurry. This difficulty is observed in Gunnarsson and Persson's project (2013) and in Mortensson and Reshetyuk's project (2015) as well.

Figure 17. Natural GCP 10. Center of a cement block.

Figure 18. Center point for GCP 3.

Figure 19. Reflectance from white background. Figure 20. Diffusion of a marker.

Reflectance from the white background dominates and wipes out the black triangles. The camera lens is thus "blinded" by the whiteness of the ground marker (Figure 19). Diffusions of the triangles

(24)

Page | 24

on the markers are also observed (Figure 20). Since the alignment horizontally and vertically and scaling are highly depended on GCPs, this greatly affects the production of dense cloud.

The size and shape of the GCP affects its correct spatial identification. Maybe the triangle form can be avoided. The formations shown in Figure 21 and Figure 22 can be more of an advantage especially when the sun is high and cloud free.

Figure 21. Suggested design of GCP (modified from Lantmäteriet's proposal).

Figure 22. Suggested GCP form (Barry and Coakley, 2013).

Another factor to consider is the color and contrast of the ground marker as against the background.

It might be preferable to use another color that does not reflect as much as white. This consideration is mentioned likewise in Gunnarsson and Persson's project (2013). The slope of the terrain can also give disadvantage in determining the central point of the marker as the marker will appear skewed (Figure 23).

Figure 23. Skewed GCP.

The produced DEM files there GCPs are not used give unanticipated results with several meters

deviation in altitude as seen when the DEM files are imported in ArcMap (Figure 24 and Figure 25). For that reason, no further analyses are made with the DEMs produced when GCPs are

not used.

(25)

Page | 25

Figure 24. DEM file, 76 meters, without GCPs. Figure 25. DEM file, 105 meters, without GCPs.

Due to the enormous difference in resolution, a few centimeters in DEMs as against two meters with Lantmäteriets Höjddata grid 2+, data are not comparable and an analysis is therefore not practical. It is therefore more realistic to make the comparison between DEMs and GNSS measured GCPs.

Unexpectedly, the results from the 105 meters flight height gives better positional accuracy than the results from 76 meters flight height with or without the use of GCPs. The RMS error at 76 meters flight height with GCPs is 0.683 meter and 0.659 meter at 105 meters flight height. The RMS error at 76 meters flight height without GCP is 2.292 meters while at 105 meters RMS error is 2.126 meters. On the other hand, DEM results give better altitude accuracy at flight height 76 meters than 105 meters with a few millimeters, with RMS error of 0.737 meter and 0.743 meter respectively. A possible explanation is that alignment is probably better accomplished at the 105 meters flight height as each image contributing to orthophoto production covers a larger area and results in a more effective point matching but this needs to be proven. In Gunnarsson and Persson's project (2013), their theory for GCP's influence on DEM is that GCPs are chosen horizontally and not vertically.

Likewise, scaling error is also observed throughout the orthophotos. Scaling is inconsequent in different areas in each orthophoto as concluded from the relative accuracy test shown in table 6 in which distances between two points are compared.

Due to the factor that photos having been taken in June where leaves have sprouted and matured and grasses have grown high, the placement of the GCPs has not been ideally done as planned. It is also observed that Photoscan fails to align images taken with tight vegetation and on water area.

(26)

Page | 26

6. Conclusion

Assignment of GCPs plays a significant role during orthophoto processing for absolute easting, northing and height position accuracy in a projected reference system. It is also important upon the production of a more correctly scaled end-product. The consequence of not using GCPs gives variety in NE deviation from GNSS NRTK measurements from a few decimeters to a few meters in orthophotos. In the absence of GCP, the elevation in DEM is greatly displaced.

However, there is no significant difference in the absolute position accuracy test between orthophotos produced with flight heights 76 and 105 meters, with or without the assignment of GCPs. Lower flight height does not guaranty better results.

Some areas and objects in the orthophotos might be skewed or lacking in data (holes). Holes are observed where leaves are concentrated and within water areas. Long vertical objects like electric poles may not satisfy the objective of a real orthophoto in the sense that residual tilts are observed.

Not all vertical features are projected correctly even if the buildings seem to have right vertical angle.

Shift/side offset of hanging power lines gives doubt to proper alignment of hanging objects. It is to be concluded that relief displacement is not completely ortho-rectified and the orthophotos are not totally free from distortions.

For future studies, it might be preferable that GCPs are placed on flat horizontal ground and at regular intervals with suitable color contrast and avoid overwhelming reflectance. The size of the marker can be set depending on the resolution applied on the camera and the feature of the camera lens. Height offset for markers is to be avoided. The flying speed of the UAV during image taking can be tested for their impact on the spatial accuracy test of the produced orthophotos. Another suggestion is in cases of difficult terrain where elevation difference is high, increase overlapping of images is recommended. GCPs placement can be tested at regular intervals both horizontally and vertically.

(27)

Page | 27 References

Agisoft LLC. (2016a). Agisoft PhotoScan User Manual Professional Edition version 1.2.

Agisoft LLC. (2016b). Dense Cloud Classification and DTM Generation with Agisoft PhotoScan Professional. Support Tutorials. Available at: http://www.agisoft.com/index.php?id=35 [Accessed 1 July 2016].

Agisoft Wiki. (2012). Agisoft PhotoScan – Capabilities. Available at:

http://www.agisoft.ru/w/index.php?title=PhotoScan/Capabilities [Accessed 1 July 2016].

Axelsson, M. and Skoog, E. (2013). Osäkerhet vid fotogrammetrisk kartering med UAS och naturliga stödpunkter.

Degree project. Högskolan i Gävle, Akademin för teknik och miljö, Avdelningen för industriell utveckling, IT och samhällsbyggnad. Available at: http://www.diva-

portal.se/smash/get/diva2:632138/FULLTEXT01.pdf [Accessed 12 June 2016].

Barry, P. and Coakley, R. (2013). Field Accuracy Test of RPAS Photogrammetry. Baseline Surveys Ltd. Available at:

downloads.agisoft.ru/articles/Barry.pdf [Accessed 12 June 2016].

Draeyer, B. and Strecha, C. (2014). White paper: How accurate are UAV surveying methods? Pix4D. Available at:

https://support.pix4d.com/hc/en-us/articles/202557229-Scientific-White-Paper-How-accurate-are- UAV-surveying-methods-#gsc.tab=0 [Accessed 13 June 2016].

Falkner, E. and Morgan, D. (2002). Aerial Mapping, Methods and Application 2nd ed. chap. 14.2. USA: CRC Press LLC. Available at http://gis-lab.info/docs/books/aerial-mapping/cr1557_14.pdf [Accessed 12 June 2016].

Graham, R. and Koh, A. (2002). Digital Aerial Survey: Theory and Practice. USA: CRC Press. p. 164. Available at:

https://books.google.se/books?isbn=0849316014 [Accessed 12 June 2016].

Gunnarsson, T. and Persson, M. (2013). Stödpunkters inverkan på osäkerheten vid georeferering av bilder tagna med UAS. Degree project. Högskolan i Gävle, Akademin för teknik och miljö, Avdelningen fär industriell utveckling, IT och samhällsbyggnad. Available at: http://hig.diva-

portal.org/smash/get/diva2:628351/FULLTEXT01.pdf [Accessed 12 June 2016].

Jansson, A. (2013). En noggrannhetsundersökning av fotogrammetrisk detalj mätning i stereo. Degree project. Karlstad Universitet, Fakulteten för humaniora och samhällsvetenskap, Naturgeografi. Available at:

http://www.diva-portal.org/smash/get/diva2:642189/FULLTEXT01.pdf [Accessed 12 June 2016].

Lantmäteriet. (2015a). HMK – handbok i mät- och kartfrågor Bilddata. Available at:

http://www.lantmateriet.se/globalassets/om-lantmateriet/var-samverkan-med-andra/handbok-mat- -och-kartfragor/hmk-geodatainsamling/2015/hmk-bilddata_2015.pdf [Accessed 12 June 2016].

Lantmäteriet. (2015b). HMK – handbok i mät- och kartfrågor Fordonsburen Laserdatainsamling. Available at:

https://www.lantmateriet.se/globalassets/om-lantmateriet/var-samverkan-med-andra/handbok- mat--och-kartfragor/hmk-geodatainsamling/2015/hmk_fordonsburen_laserdatainsamling_2015.pdf [Accessed 12 June 2016].

Lantmäteriet. (2015c). HMK – handbok i mät- och kartfrågor Laserdata. Available at:

https://www.lantmateriet.se/globalassets/om-lantmateriet/var-samverkan-med-andra/handbok- mat--och-kartfragor/hmk-geodatainsamling/2015/hmk-laserdata_2015.pdf [Accessed 12 June 2016].

(28)

Page | 28

Lantmäteriet. (2015d). HMK – handbok i mät- och kartfrågor Ortofoto. Available at:

https://www.lantmateriet.se/globalassets/om-lantmateriet/var-samverkan-med-andra/handbok- mat--och-kartfragor/hmk-geodatainsamling/2015/hmk-ortofoto_2015.pdf [Accessed 12 June 2016].

Lantmäteriet, Kungliga tekniska högskolan, Lunds universitet and Högskolan i Gävle (2013). Geodetisk och fotogrammetrisk mätnings- och beräkningsteknik, Kapitel 13-15. Available at:

https://www.lantmateriet.se/globalassets/om-lantmateriet/var-samverkan-med-andra/handbok- mat--och-kartfragor/utbildning/kompendium_131028_kap13-15.pdf [Accessed 12 June 2016].

Mortensson, S-G. and Reshetyuk, Y. (2015). Noggrann och kostnadseffektiv uppdatering av DTM med UAS för BIM.

Trafikverket. Available at: http://hig.diva-

portal.org/smash/record.jsf?pid=diva2%3A920253&dswid=8466 [Accessed 1 July 2016].

Orthophoto. (n.d.). In: ESRI support GIS dictionary. Available at:

http://support.esri.com/sitecore/content/support/Home/other-resources/gis- dictionary/term/orthophoto. [Accessed 12 June 2016].

Persson, C-G. (2013). HMK – handbok i mät- och kartfrågor. Lägesosäkerhet vid fotogrammetrisk detaljmätning i 3D, Teknisk rapport. Lantmäteriet. Available at: http://www.lantmateriet.se/globalassets/om-

lantmateriet/var-samverkan-med-andra/handbok-mat--och-kartfragor/tekn_rapporter/hmk_2013- 3.pdf [Accessed 12 June 2016].

Pix4D support site. (2016a). Can Pix4Dmapper process images taken with different flight heights? Available at:

https://support.pix4d.com/hc/en-us/articles/202558979-Can-Pix4Dmapper-process-images-taken- with-different-flight-heights-#gsc.tab=0 [Accessed 12 June 2016].

Pix4D support site. (2016b) How to verify that there is enough overlap between the images. Available at:

https://support.pix4d.com/hc/en-us/articles/203756125-How-to-verify-that-there-is-Enough- Overlap-between-the-Images#gsc.tab=0 [Accessed 12 June 2016].

Pix4D support site. (2016c). Step-1-Before-Starting-a-Project-1-Designing-the-Image-Acquisition-Plan-b-Computing-the- Flight-Height-for-a-given-GSD. Available at: https://support.pix4d.com/hc/en-us/articles/202557469- Step-1-Before-Starting-a-Project-1-Designing-the-Image-Acquisition-Plan-b-Computing-the-Flight- Height-for-a-given-GSD#gsc.tab=0 [Accessed 12 June 2016].

Pix4D support site. (2016d). Step 1. Before Starting a Project > 4. Getting GCPs on the field or through other sources (optional but recommended). Available at: https://support.pix4d.com/hc/en-us/articles/202557489-Step- 1-Before-Starting-a-Project-4-Getting-GCPs-on-the-field-or-through-other-sources-optional-but- recommended-#gsc.tab=0 [Accessed 12 June 2016].

Pix4D support site. (2016e). Tools GSD Calculator. Available at: https://support.pix4d.com/hc/en- us/articles/202560249-TOOLS-GSD-Calculator#gsc.tab=0 [Accessed 12 June 2016].

Sveriges meteorologiska och hydrologiska institut [SMHI]. (2015). Vad är ett solbanediagram? Available at:

http://www.smhi.se/polopoly_fs/1.32097.1398236573!/image/stockholm_kontur.png_gen/derivati ves/Original/stockholm_kontur.png [Accessed 12 June 2016].

Samani, J. (2013). UAS-noggrannhet i praktiken – En undersökning av dagens UAS-fotogrammetris noggrannhet. Degree project. Karlstads universitet, Fakulteten för humaniora och samhällsvetenskap, Naturgeografi.

Available at: http://kau.diva-portal.org/smash/get/diva2:691154/FULLTEXT01.pdf [Accessed 12 June 2016].

(29)

Page | 29

Semyonov, D. (2011) Algorithms used in Photoscan. Community Forum, Agisoft LLC. Available at:

http://www.agisoft.com/forum/index.php?topic=89.0 [Accessed 1 July 2016].

U.S. Geological Survey [USGS]. (2013). Digital orthophotos. Available at:

http://online.wr.usgs.gov/ngpo/doq/doq_basics.html [Accessed 12 June 2016].

Verbyla, D. L. (1995). Satellite Remote Sensing of Natural Resouces. USA: CRC Press LLC. Available at:

https://books.google.se/books/about/Satellite_Remote_Sensing_of_Natural_Reso.html?id=Ps0Kx zqHdlYC&redir_esc=y [Accessed 12 June 2016].

Verhoeven, G. (2011). Taking computer vision aloft – archaeological three-dimensional reconstructions from aerial photographs with photoscan. Archaeological Prospection, 18(1), 67–73, January/March 2011 DOI:

10.1002/arp.399. John Wiley & Sons, Ltd. Available at:

http://www.atelier3d.levillage.org/IMG/pdf/VERHOEVEN_2011_Taking_Computer_Vision_Alo ft_Archaeological_3d_reconstruction_from_Photoscan.pdf [Accessed 1 July 2016].

(30)

Page | 30

Appendix

Appendix 1 1. Camera and Photoscan processing parameters

General settings used in this project paper:

(31)

Page | 31

Appendix 2 2. Coordinates of the ground markers

(32)

Page | 32

Appendix 3 3. System requirements for Photoscan

For system requirements can be found at the Agisoft homepage http://www.agisoft.com/downloads/system-requirements/.

RAM

In most cases the maximum project size that can be processed is limited by the amount of RAM available. Therefore it is important to select the platform allowing installing required amount of RAM.

See Memory Requirements on Agisoft website.

(http://www.agisoft.com/pdf/tips_and_tricks/PhotoScan_Memory_Requirements.pdf).

CPU

Complex geometry reconstruction algorithms need a lot of computational resources for processing.

A high speed multi core CPU (3GHz+) is recommended.

GPU

Agisoft PhotoScan supports OpenCL acceleration for dense cloud generation step (most time consuming one), so high-end OpenCL-compatible graphics card can speed up the processing.

(33)

Page | 33

Appendix 4 4. Processing steps in Photoscan

General settings used in orthophoto and 3D model processing.

Step 1. Photo alignment

Agisoft Photoscan uses common points on the photographs to match images, aligning these images and calculate respective camera positions for each image. Alignment is done with the SFM, structure for motion, technique. SFM detects geometrical similarities with specific details that serve as image feature points. The movement of these points throughout the whole sequence in the workplace is thereby monitored giving an estimation of feature point positions and subsequently rendered as three-dimensional point cloud. When these are identified, Photoscan refines camera calibration parameters to create point cloud data and a set of camera positions together with a list for alignment errors. (Verhoeven, 2011).

At this step, point variance is calculated and three sigma filtering is applied before SFM algorithms are applied. Camera adjustment is refined using a bundle-adjustment algorithm (Semyonov, 2011).

105 Flight height, alignment result 76 Flight height, alignment result

(34)

Page | 34

Appendix 4

Step 2. Set markers

At this point GCPs are imported for georeferencing and their positions are identified on images where applicable. Conversion as required depending on the reference system used in the images are set inline with the reference system used by the GCPs (GCPs estimated vs. measured data error).

This information can be used to check whether the result fulfills one's project requirements for the spatial and georeferencing accuracy.

Before conversion After conversion

Step 3. Building the dense cloud

Dense reconstruction algorithms are applied on the aligned image set operating on the pixel values to build a geometric scene.

(35)

Page | 35

Appendix 4

Step 4. Building mesh

Multiview stereo reconstruction handles fine details present on the scene and represent these details as a mesh. Height field approach is recommended as default for aerial photography in the reconstruction of terrain-like features (Verhoeven, 2011). The terrain model is a result of the triangulation of the points in the mesh .

Apply topology fix when applicable on the resulting mesh.

Step 5. Building texture

The mapping mode orthophoto is used in processing nearly planar geometry. This mapping method implements an adaptive parameterization approach in Photoscan. Horizontal surfaces are mapped with orthophoto parameterization mode while vertical regions are mapped using generic mapping mode when building texture. This option is therefore recommended in processing aerial photographs.

(36)

Page | 36

Appendix 4

Step 6. Building DEM and orthomosaic

Photoscan uses the TIN surface to correct for terrain displacement and calculated exterior orientations for georeferencing in the orthorectification process (Agisoft Wiki, 2012). The surface reconstruction algorithm is based on a modified Poisson surface reconstruction algorithm (Semyonov, 2011).

1. 105 m flight height with GCP

(37)

Page | 37

Appendix 4

(38)

Page | 38

Appendix 4

2. 105 m flight height without GCP

(39)

Page | 39

Appendix 4

(40)

Page | 40

Appendix 4

.

(41)

Page | 41

Appendix 4

Chunks summary:

(42)

Page | 42

Appendix 5 5. Computation tables

(43)

Page | 43

Appendix 5

(44)

Page | 44

Appendix 5

Relative spatial accuracy test

(45)

Page | 45

Appendix 6 6. Orthophoto produced from images taken at 105 m flight height and with

assignment of 10 GCPs.

(46)

Page | 46

Appendix 7 7. GCPs' position comparison

7.1 GNSS measured vs. 105 m flight height and 76 m flight height, with GCP.

(47)

Page | 47

7.2 GNSS measured vs. 105 m flight height and 76 m flight height, without GCP.

Spatial Accuracy in Orthophoto produced using UAV Photographic Images