UAV data acquisition and processing

Transcription

1 Deliverable D.2.02 UAV data acquisition and processing WP 2 Forest information collection and analysis Task UAV data acquisition and processing Revision: Final Authors: Enda Nolan Participant: Coastway Surveys Dissemination level Contributor(s) Reviewer(s) PU (Public) Enda Nolan (Coastway), David O Reilly Liam Murphy (Coastway), Federico Prandi Editor(s) Partner in charge(s) Due date Submission Date Raffaele De Amicis (Graphitech) FLY, CNR, COAST,TRE 31-MAR APR-15 Page 1 of 127

2 REVISION HISTORY AND STATEMENT OF ORIGINALITY Revision History Revision Date Author Organisation Description /07/2014 Enda Nolan Coastway Surveys Definition of the document Chapters /10/2014 Enda Nolan Coastway Surveys Content and Revision /12/2014 Enda Nolan Coastway Surveys Revision /03/2015 Federico Prandi Graphitech Revision Statement of originality This deliverable contains original unpublished work except where clearly indicated otherwise. Acknowledgement of previously published material and of the work of others has been made through appropriate citation, quotation or both. Page 2 of 127

3 Table of contents REVISION HISTORY AND STATEMENT OF ORIGINALITY... 2 Revision History... 2 Statement of originality... 2 Table of contents... 3 List of figures... 5 Acronyms Introduction Overview of UAV data collection for forestry applications Payloads: Cameras and sensors Processing and Outcomes Regulation & Licencing Pilot Test Sites Test Site Locations Ireland & Italy Phase 1 (Phase 2 Austria in 2015) Gortahile Trento Italy (site 2) Survey Equipment UAV Image Sensors IXUS/ELPH RGB S110 NIR Standard GNSS GEO 7X Flight Methodology Pre Flight Checks Flight Post-Flight Checks and Comments Processing and Outputs Processing Chain Software Tools Adopted Operations and Processing Methodology Data Outputs Additional Outputs Summary of Survey Results Piscine (RGB) Piscine (NIR) Montesover (RGB) Montesover (NIR) Page 3 of 127

4 8 Annex Quality Report Outputs Checklist Bibliography Page 4 of 127

5 List of figures Figure 1: Fixed wing e-bee UAV sistem... 8 Figure 3 Flight plan Figure 4 Flight plan Figure 5 UAV transport box Figure 6 ebee UAV system Figure 8 S110 NIR camera Figure 9 Trimble geo 7x Figure 10 Flight path Ireland Figure 12 Process chain Figure 13 New project Figure 15 Import images Figure 19 Window to locate the GCPs Figure 20 Colored point cloud obtained for the test site on Italy Figure 22 Undistorted image Figure 23 Low resolution ortomosaic Figure 25 Dense matching raster DSM Figure 26 Dense matching grid point DSM Page 5 of 127

6 Acronym CA DM GA GA OM PC QRM SB TB TL WPL DTM SqKm RPAS IAA CAA EC Consortium Agreement Data Manager Grant Agreement General Assembly Operational Manager Project Coordinator Quality and Risk Manager Stakeholders Board Technical Board Task Leader Work Package Leader Digital Terrain Model Square Kilometre Remotely Piloted Aerial Systems Irish Aviation Authority Civil Aviation Authority Page 6 of 127

7 1 Introduction Precision Agriculture and forestry applications require access to high spatial and temporal resolution and up to date images. Using traditional techniques such as satellite and airborne remote sensing data this could be difficult and expensive. The use of Airborne Lidar over the last 10 to 15 years has proven advantageous but limited in its capability to create a DTM due to the varied level of undergrowth, leading to the creation of a very sparse DTM. Advances in the last five years in the capabilities of lightweight UAV/UAS equipment and payloads for civilian use have brought massive improvements in the methods used to survey large land parcels agriculture including forestry. These systems provide a potential alternative to satellite and airborne data acquisition ensuring low cost, high spatial and temporal resolution, and high flexibility in image acquisition programming. Lightweight battery powered UAV s have proven to be a great leap forward and an excellent cost effective Surveyors tool for land flat parcels up to 10sqkm, which do not need extensive clearing of scrum and undergrowth to determine the DTM, the Slope project will determine the true value of this equipment in mountainous areas. In an agricultural or forestry scenario, in order to produce valid photogrammetric outputs, the fixed wing UAV with autonomous navigation system uses GPS and inertial measuring technology which are the preferred methods with respect to the RPAS systems currently available. The SLOPE projects intention is to investigate the application of UAV systems for locating harvesting sites, planning and supporting forestry operations. The extent of the area of interest is in general relatively small (less than 1500 ha) thus it is required to obtain data and analysis in a relatively short time. Furthermore the positional accuracy is generally low (< 1m) and a visual analysis of the images is important so the availability of high-resolution orthophotos is quite important. This document will analyse the use of a UAV system to collect and collate information to support forest harvest planning in mountainous areas. In particular the field operations will be investigated focusing on the collection of raw data, the generation of the DSM and orthophotos and the general evaluation of the accuracies achieved. Furthermore a section highlighting the pros and cons of the overall workflow will be provided. Page 7 of 127

8 2 Overview of UAV data collection for forestry applications Unmanned aerial vehicles (UAV) are becoming a feasible and low cost method for remote-sensing data acquisition, mostly aerial images and derived products such as DSM and orthophotos. Thanks to the development, reliability and miniaturization of the sensors modern UAVs systems enable a relative safe operation. However their application on forestry and mountain environment pose some challenges that will be face off during SLOPE projects, first of all the weather conditions (wind) but also the GNSS signal quality or the visibility of the terrain under the canopy coverage in absence DTM availability. Despite these limitations the potentiality of UAVs on forestry applications in terms of spatial and temporal resolution and costs paves the way to the wide usage of these instruments especially in case like mountain forest harvesting planning where precision farming could be very useful to exploit the potential of the wood preserving the natural resource. Figure 1: Fixed wing e-bee UAV sistem One of the possible classifications of UAVs system is based on the different type of flying vehicles able to transport camera and/or other sensors. There are the fixed wing UAVs and multi rotor helicopter; both systems can be piloted by an operator via Remote Control (RC) or by an on-board autopilot. In case of Auto piloted UAV the mission is planned in advance defining path, height velocity and photo trigger. The vehicle is then navigated by on-board GNSS/INS unit, in case of problem like signal loss the system switch on the semiautomatic mode and for these reasons the flight has to be controlled by a qualified pilot able to take the control in every moment. In case of forestry application and precision farming when the area to be covered is relatively big and there is not particular requirements to survey on vertical surfaces the fixed wing system are preferred. Page 8 of 127

9 2.1 Payloads: Cameras and sensors Due their nature small commercial UAVs (especially fixed wing ones) have reduced payload and space, for these reasons these systems are typically equipped with compact consumer cameras. The resolution of these cameras can be very high ensuring a high spatial resolution of the acquired images, however some problems related to the stability of the inner orientation parameters as well as the rolling shutter effect can be affect the final product. For agricultural and forestry applications other sensors like multi/hyperspectral ones can be mounted on the UAVs for vegetation analysis. 2.2 Processing and Outcomes The first outcome of the UAV forest images is the production of Digital Surface Model (DSM) obtained by image matching and the related orthorectified image mosaic. Thanks to the high-resolution aerial images acquired this products can became the basis for subsequent geospatial analysis. Within SLOPE particular attention will be dedicated to the in single-tree-extraction (Cfr. D2.3), tree crown detection, tree density, age-classes and further structure parameters, like stem or timber volume. These parameters can be derived either from object-based image analysis or from point cloud data as a result of image matching. Forest parameters estimated using UAV derived information and refined by the other SLOPE survey will be a valuable support for the forestry inventory and the forest information system at the base of SLOPE project. 2.3 Regulation & Licencing The absence of an EU wide regulatory framework limits the possibility to fly UAV s in European airspace. It is a severe limitation for the development of UAV market, which requires careful consideration. A pilot project carried out in 2009 has set in place developments which will eventually lead to a single regulation and licencing agreement for UAV s / RPAS. At the beginning of the Slope Project each test site has its own regulatory body, Ireland IAA Irish Aviation Authority, Italy - ENAC Italian Civil Aviation Authority, Austria Austro Control. Page 9 of 127

10 Figure 2: Foreseen roadmap for UAV EU regulamentation Every private operator must obtain a licence to fly from several of the Aviation Authorities in Europe, Coastway Surveys is licenced by the CAA & IAA. Page 10 of 127

11 3 Pilot Test Sites In order to apply the UAV survey acquisition techniques and to provide feasible input for the whole SLOPE supply chain, some data acquisition campaigns have been planned and realised. Two test survey have been realized during the first project year in Ireland and Italy, a second campaign will be realized during the second year in Austria in order to apply the lesson learned during the previous ones, and to define the final SLOPE methodology. 3.1 Test Site Locations Ireland & Italy Phase 1 (Phase 2 Austria in 2015) Gortahile Located in County Laois Ireland at approximately 400m above sea level this area was chosen to test the UAV and carry out a site calibration of the payload cameras. The test site consists of a stand of forest 3 hectares in size consisting of a single species Norwegian Spruce with stem diameters ranging from mm. The site was chosen, as there was clear access to the forest to allow laser scanning to be carried out in conjunction with UAV flights and there is good GPS coverage. Figure 3 Flight plan Irish Forest Survey Mission Planning In a simulation of the survey we used a combination of traditional GPS surveying & Laser Scanning and Aerial Mapping using the UAV and Faro scanner. Surveyors placed ground targets around the forest and recorded the GPS coordinates of each point; several Surveyors commenced laser-scanning stands of trees approximately 100mx100m in size to help create a digital terrain model (DTM), a Digital Surface Model (DSM) and a digital canopy model (DCM). Page 11 of 127

12 The UAV Pilot and Commander made their way to a mountain top located approximately 1km from the laser scanning sites, communications were maintained by radio at all times. UAV control software (emotion2) was used by the pilot to locate the forest stand on a tablet device. The recorded coordinates of the Targets were entered into the software and a safety flight buffer zone placed approximately 500m around the outside of the survey area. The safety buffer zone is used as a return to base barrier should the UAV be taken off course by the wind. Wind Speed to be taken into account when scanning if above 10km at ground level blurring will occur in point cloud. The wind at test flight time was over 30km per hour so the pilot was placed on standby as he did not receive flight clearance from Air Traffic Control in Dublin. The wind dropped, all Surveyors and Spotters were informed by the Commander that the flight was about to begin. Take off, flight duration 45minutes. Several test flights emergency landings/loss of control/ and find the UAV scenarios were practiced Trento Italy (site 2) This location was chosen as a suitable test site for the installation of choker head cable routes. The difference in elevation between the highest point of the site and the lowest is approximately 450 metres; the total area is 36 hectares. The site consisted of forestry of mixed species with stem diameters up to 800mm and a tree crown elevation of 30 to 40 metres. Figure 4 Flight plan The Coastway team of three key personnel was installed to ensure the correct variation of skills require from UAV Pilot, Commander and Surveyor. Page 12 of 127

13 All equipment is easily transported by domestic flights between countries, making this method of surveying mobile and cost effective in comparison to moving Lidar mounted aeroplanes between countries. Coastway contacted Slope partners Graphitech, CNR & Treemetrics to aid in local translation and staff accommodation. The purpose of the test flights was to calibrate equipment in the forest environment, this included the UAV, GPS, Canon RGB Camera, Canon Near Infrared camera, manufacturers calibration certificates for all equipment is included in Appendix A of this document. Field test calibrations were carried out using Coastway s 1m x 1m targets placed on a flat surface such as a road, the target has a white cross which can be viewed in Image 19 below. A GPS reading is taken with the Trimble Geo7X placed on the target; eastings, northings and level are recorded. The UAV is then flown over the target and set to record imagery at 80% overlap. Post processing of the data revealed the eastings, northings and level of the data produced by the UAV. Figure 5 UAV transport box Test Flight Planning The project coordinator for the test case provided approximate coordinates of the cable lines. On the basis of this and the terrain morphology a flight plans were drafted on emotion software for discussion, which will be adjusted to suit when the survey crew arrives on site. To acquire usable data it is necessary to plan a ground layout of targets which should for best practice purposes be placed at aproximately 300m centres and must be clearly visible from above. This involves visiting the survey site and carrying out reconnasiance and cannot be substituted by viewing digital sattelite maps which can be outdated by years and will not have up to date growth or changes to the physical location. Page 13 of 127

14 Planning a test mission will involve contacting the relative Aviation Control Authority and seeking permission to fly within their airspace. Controlled airspace may be imposed on areas due to military, environmental, weather conditions, privacy regulations the list is ever growing and contact must be made to seek permission to fly. Coastway wrote to ENAC in Italy in early June 2014 and passed on the request to letter to the the Slope Coordinator & WP member CNR who wrote to ENAC on behalf of the project. The request to fly clearly states that we represent the Slope Project and our project is a research project funded by the EC for forestry management purposes. Page 14 of 127

15 4 Survey Equipment 4.1 UAV The UAV chosen is a small unmanned aircraft weighing less than 700g. Manufactured by Sensefly, the ebee was chosen due to its compact nature, camera options, onboard sensors and semi-autonomous capabilities. The ebee is hand launched, controlled and landed with the aid of an autopilot system which uses data provided by the Inertial Measurement Unit (IMU) and onboard GPS as well as other on-board sensors. No manual intervention is needed, but an option has been provided to enable the user to take full manual control if desired. The UAVs characteristics are: 96cm wingspan Less 0.55kg dry-weight (0.68kg with RGB payload, 0.71kg with NIR payload) Lithium Polymer battery (11.1 V, 1800 mah) minute flight time km/h cruise speed Up to 45km/h or 12m/s wind resistance Up to 3Km radio link Up to 12sqkm coverage Linear landing Image resolution/pixels of 3-30cm Autopilot Flight Planning (emotion 2) EPP foam, carbon structure & composite parts Electric pusher propeller, 160 W brushless DC motor Figure 6 ebee UAV system Page 15 of 127

16 4.2 Image Sensors The sensors used for the surveys are the ones provide as default with the ebee UAVs system. They are: a commercial RGB camera and a Near InfraRed camera designed for agricultural applications IXUS/ELPH RGB This customised 16 MP camera is electronically integrated within the ebee s autopilot. The 127HS RGB acquires regular image data in the visible spectrum. Example applications: real colour 2D and 3D visual rendering. Resolution 16MP 35 mm equivalent focal length 24mm Flight altitude (4cm/px GSD) Sensor size Pixel pitch Image format Settings 130 m (optimal conditions) 6.16mm x 4.62mm 1.33 μm JPEG Only automatic Figure 7 IXUS/ELPH RGB camera Page 16 of 127

17 4.2.2 Flight Elevation m GSD cm/px (optimal conditions) Single shot coverage (Width x Height) m x 140 = m² x 172 = m² x 204 = m² x 236 = m² x 269 = m² x 301 = m² x 338 = m² S110 NIR Standard The S110 NIR acquires image data in the near infrared (NIR) band, the region where high plant reflectance occurs. Its exposure parameters can be set manually and its RAW files are fully supported by the ebee Ag s software. Example applications: biomass indication, growth monitoring, crop discrimination, leaf area indexing. This customised 12 MP camera is electronically integrated within the ebee s autopilot. Resolution 12MP 35 mm equivalent focal length 24mm Flight altitude (4cm/px GSD) 115 m (optimal conditions) Page 17 of 127

18 Sensor size Pixel pitch Image format Settings 7.44 x 5.58 mm 1.86 μm JPEG and RAW Automatic and Manual Figure 8 S110 NIR camera 4.3 GNSS GEO 7X A GPS is needed in order to obtain the GCP coordinates of the Groud control points. The chosen GPS is the GEO 7X, due to its compact nature and support for all GNSS networks, and technologies allowing for the reduction of multipath errors. It s IP65 and MIL-STD-810F ratings and all day battery makes it suitable for forest conditions. System Summary Dual-frequency GNSS receiver and antenna Sunlight readable 4.2 polarized display Integrated 3.5G cellular modem Integrated Wi-Fi and Bluetooth wireless technology 5 megapixel autofocus camera Microsoft Windows Embedded Handheld version 6.5 Professional. Rugged and water-resistant design Performance Specifications Code differential GNSS positioning Horizontal m + 1 ppm RMS Page 18 of 127

19 Vertical m + 1 ppm RMS SBAS differential positioning - typically <5 m 3DRMS Real-Time Kinematic surveying Single Baseline <30 km o Horizontal - 25 mm ppm RMS o Vertical - 40 mm ppm RMS Network RTK o Horizontal - 25 mm + 1 ppm RMS o Vertical - 40 mm + 1 ppm RMS Initialization time - typically <8 seconds Initialization reliability - typically >99 9% Figure 9 Trimble geo 7x Page 19 of 127

20 5 Flight Methodology In this section are described the practical operation to be executed on field in order to perform a good UAV survey of a forest plot. The UAV surveys are a quite new techniques to acquire data therefore the methodologies and the workflows to be followed are not yet consolidate, if compared to the typical aerial or satellite campaigns acquisition. 5.1 Pre Flight Checks As with any aircraft, pre-flight checks were carried out to determine the safety of the aircraft. The checks are in line with the manufacturers brief and slope project flight manual. See appendices. Coastway contacted the manufacturer based in Switzerland to be on standby should the equipment be damaged: extra wings and batteries were brought on the test case. The UAV is fitted with a tracking device which Coastway can use to track the equipment for several days after loss. Bird of prey attack is a possibility. Raptors are territorial the manufacturer has installed a safe bird deterrent software. Coastway planned numerous flights to ensure full coverage and data capture using RGB and NIR cameras. Flight plans are created using the emotion 2 software. Due to the naturally complex visual patterns of forestry, generally a flight plan with high overlap (70% frontal overlap and 85% side overlap), with perpendicular flight lines at a height of 150m is employed. Mountainous areas often have their localized meteorological conditions, with different wind speed in valleys and over the canopy. The UAV manufacturer stipulates no flights during rain as this will most likely render the data unusable. 5.2 Flight Once a full site visit has been completed by the survey crew a full flight plan can be drawn up. The flight plan must adhere to strict regulation detailing the legal altitude at which the UAV may fly above the ground, this varies for different countries but is normally 150m maximum. The UAV must at all times be within eye sight of the operator or a designated spotter. The testing carried out at Gortahile followed this procedure and the flight plan was drafted and uploaded to the autopilot of the UAV. The pilot set the maximum altitude to 100m and noted the number of waypoints the UAV would have to travel to ensure full coverage of the site. Page 20 of 127

21 Figure 10 Flight path Ireland Figure 11 Flight path Italy site 2 Prior to flights markers shall be set out around the site to act as ground control points (GCPs), a minimum of 3 GCPs should be used, with that number increasing with the size of the site. GCPS should be placed in a grid pattern, if possible covering the whole surveyed area, this helps with redundancy therefore influencing the georeferencing of the project. GCP positions are then recorded using GPS, the position of GCPs should be recorded as accurately as possible. GCPs are left in position for the duration of the survey to ensure accuracy of all data during processing. If ground control is used, final data quality can be improved, however, GCPs are not required, but are recommended to improve the geolocation and completeness of the reconstruction. Expected resolution is between 4-8 cm/pixel GSD given optimal conditions. When GCPs have been set out, the flight can begin and the surveyors get into the preliminary position for take-off. The take-off and landing areas are refined pre-takeoff and the complete flight plan is uploaded to the UAV. A minimum crew of 2 is needed, a spotter and a pilot. The spotter ensures that the UAV is following the designated flight plan, and will alert the pilot if the UAV gets into difficulty. Page 21 of 127

22 5.3 Post-Flight Checks and Comments All datasets were combined with the ultimate goal of creating a 3D model of the forest: DTM-DSM- Canopy model, orthomosaic, 3D point cloud, and vegetation index calculations. Data was delivered to the next slope partner who shall use the data to create models. The following table contains comments and highlight issues encountered on the specific SLOPE test case: Survey of a forest plot in steep mountain environmental. General feedback and issues encountered Accuracy of GCPs: Due to the topography of the site, GNSS coverage of the area was poor on the ground. Coastway would recommend using an ebee RTK to ensure good georeferencing. Satellite availability while in flight: The minimum of 4 satellites for the handheld GPS and the minimum of 6 for the UAV were not always available due to the mountainous topography and heavy forest cover blocking satellite signals from the receivers (Multipath errors). This created issues with the collection of GCPs and ground truth. Localized meteorological conditions: Intermittent fog disrupted flight time available, as very little usable data can be obtained due to visual obstruction. Back winds created by the topography of the site also created issues mid-flight, as the UAV had to correct itself semifrequently therefore using more battery power. Landing: Ground Sensor should be disabled in sloped landing areas in order to ensure safe landing. During one flight, with the ground sensor on, the ebee detected terrain/canopy during the landing procedure (the ebee assumes landing site is relatively flat) and decided it was safe to descend. Due to the slope of the site, the angle of the linear landing, and the loss of 2 satellites diminishing the overall positional accuracy, the UAV descended into the canopy. Battery drain during 3D flight: When the UAV was following a 3D flight plan, it had to rapidly ascend and descend multiple time creating high levels of battery drain. Payload issues: Although the weight difference between the two payload options is only 0.03kg, flying with the NIR camera reduces flight time by 5-10 minutes, this caused an issue when landing as the height was set to 100m for the home waypoint to give extra clearance from the canopy. The landing started after 30 minutes of flight time. Page 22 of 127

23 During the linear landing, the ebee stayed 3 minutes on the home waypoint and it increased the altitude 3 times up to 100m. The battery was completely drained after this flight. Summary of factors affecting final accuracy: GPS was the main factor in issues created; multipath errors were extremely prevalent due to the topology of the site. These issues were anticipated, but not to the extent that they occurred. Weather Conditions: Data captured during one flight had significant cloud/fog cover, diminishing the usability of the data. Combined datasets: Using the geometry from obtained from both RGB and NIR datasets improved the accuracy of the georectification (see quality reports) The Pros & Cons of methodology: Pros - Large areas of forest can be covered in a short period of time. - An accurate DSM, DTM & DCM can be created in most instances where a clear view of the forest floor is available, this was not possible in Trento sites 1 and 2. - Areas that are inaccessible on foot are reachable to help plan access routes and harvest landings - Accurate maps allow cable line planning - Species determination can be determined by trained professionals. - Stock take, the forester can determine which areas to thin and harvest - By combining aerial data with laser scan data a forest warehouse can be created - Cost effective - Up to date records can be created and maintained Cons - Methodology requires GCPs and even with the most up to date survey equipment we were unable to acquire good quality positions for the GCPs because of multipath errors created by the topography of the site (High angle of slope). Optimal GCP placement was unobtainable due to the topology of the site (Trees/vegetation). - Fixed wing UAV requires a linear landing, which can be problematic if survey area has no suitable take off/landing sites. Page 23 of 127

24 Recommendations: Since Slope project commenced UAV technology and software have evolved. An RTK UAV with a ground control station will improve the accuracy of the survey by removing the necessity for GCPs on the ground. Thus removing the need for GPS coverage on the GCPs, this will also reduce time on site and complexity of the survey methodology. Other available camera sensors should be tested, such as the RGB S110, multispec 4C (Multispectral), and possibly the upcoming thermomap (Thermal Imaging). A lightweight LIDAR system, such as the Velodyne VLP-16 (600g) could be used on a hexacopter to penetrate vegetation/canopy to derive upto-date DTM information. Page 24 of 127

25 6 Processing and Outputs 6.1 Processing Chain In this section is described the whole process chain, including a summary of the on preparation and the execution of the survey (described on detail on chapter 6) and the processing operation in order to obtain the final products from the raw data acquired on field. Figure 12 Process chain 6.2 Software Tools Adopted Emotion 2 o Used for flight planning, and preliminary post-flight processing (Geotagging, Yaw/Pitch/Roll). Postflight Terra 3D o Used for Data Processing (Point Cloud Generation, Orthophotos, DSM, etc) Page 25 of 127

26 6.3 Operations and Processing Methodology 1) Prepare Flight Plan and Site a. Optional but recommended: Set out and Measure GCPs b. Use emotion2 for flight planning i. Set parameters for over 70% lateral and 80% Longitudinal Overlap ii. Fly over constant height (optionally use a 3D flight plan with heights obtained from the STRM dataset to improve GSD) iii. Use Perpendicular Flight Lines iv. Set camera parameters pitch angle to 3 v. (Optional) Deactivate ground sensor when working in sloped areas. vi. Upload flight plan to UAV and proceed to take-off 2) UAV flight and Data acquisition a. Launch UAV into a light breeze with no obstructions on the take-off flight line. b. Ensure a spotter always has eyes on UAV c. Spotter should confirm with pilot that UAV is following designated flight path. d. When UAV has completed data acquisition it will start landing procedures, for sloped areas, ensure that the ground sensor has been deactivated when coming in for landing, as the UAV will dive as soon as it senses land/canopy. e. Ensure that UAV is following correct landing procedures, be prepared to abort landing if UAV is not following landing procedures or is likely to hit an obstruction. If UAV is following correct landing procedures, finalise the landing and ensure that the survey team moves out of the landing flight line. 2) Rapid Processing (optional) a. While on site, transfer data to workstation, proceed with rapid processing, this will generate a rapid report which ensures that the required area has sufficient data, if rapid processing reports that there has not been sufficient data captured, repeat the flight and acquire more images. 3) Full Processing b. Create new project in Postflight Terra 3d/pix4d Figure 13 New project Page 26 of 127

27 c. Select type and Name of the Project Figure 14 Type and name of project In this case the type of the project is Aerial nadir, because the geometry of the acquisition is similar to the typical aerial photogrammetric acquisition. d. Import images Figure 15 Import images Page 27 of 127

28 e. Ensure Image Properties are correct Figure 16 Summary of loaded images information In this step the image with the approximate external orientation parameters recorded during the flight are presented. Page 28 of 127

29 f. Set Processing Options Figure 17 Processing options 1. Initial Processing Parameters - Aerial Nadir - Feature Extraction Original Image Size - Optimise externals and all internals Rematch images 2. Start Rapid check, this builds a sparse point cloud which can aid placement of ground control points Page 29 of 127

30 g. Import GCPs and select output coordinate system. Locate GCPs in images and tag accordingly Figure 18 Import GPPs Figure 19 Window to locate the GCPs h. After GCPs have been Selected, Start point Cloud densification and DSM/Orthomosaic Generation, parameters should be changed based on the desired output and requirements. Page 30 of 127

31 i. All Outputs are automatically generated during each processing stage, after full processing the data is ready to use for analysis. Figure 20 Colored point cloud obtained for the test site on Italy Figure 21 Ortomosaic obtained over imposed to the already available DTM Page 31 of 127

32 6.4 Data Outputs Postflight Terra 3D (Sensefly, 2014) Initial Processing Outputs: Camera Internals and Externals, AAT, BBA: When this option is selected, the results of the AAT, BBA, and optimized internal and external camera parameters are saved into files. Undistorted Images: This feature generates an undistorted copy of each original image using the optimized distortion parameters of the selected camera model. Figure 22 Undistorted image Low Resolution (8 GSD) Orthomosaic: This feature generates a low spatial resolution DSM and orthomosaic (in GeoTIFF format). A merged file as well as tiles are generated for both outputs. This DSM and orthomosaic have a GSD that is equal to 8 times the GSD of the original images. Page 32 of 127

33 SLOPE - Integrated processing and control systems for sustainable forest Production in Figure 23 Low resolution ortomosaic Point Cloud Outputs: Figure 24 Point cloud This section allows the user to select the desired file format for the densified and filtered point clouds. The following formats can be chosen: PLY: PLY file with X,Y,Z position and colour information for each point of the point cloud. XYZ: ASCII text file with the X,Y,Z position of each point of the point cloud. This point cloud does not have colour information. o Delimiter: Defines the delimiter character of the file, used to separate the values. The drop down list has the following options: Space Tab Page 33 of 127

34 Comma Semicolon LAS (default): LiDAR LAS file with X,Y,Z position and colour information for each point of the point cloud. LAZ: Compressed LiDAR LAS file with X,Y,Z position and colour information for each point of the point cloud. DSM and Orthomosaic Outputs: 1. Raster DSM : This section allows the user to save the DSM in the following format: GeoTIFF (activated by default): Saves the DSM into a GeoTIFF file. The resolution of the DSM is selected in the Local Processing menu of the main window. For most of the projects the DSM is split into several tiles and one GeoTIFF file is generated per tile. Merge Tiles (activated by default): Generates a single DSM GeoTIFF file by merging the individual tiles. When this option is deactivated, the merged DSM file is not generated. Figure 25 Dense matching raster DSM 2. Grid DSM : This section allows the user to select the desired file format for the vector DSM: XYZ: LAS: LAZ: Spacing [cm]: The spacing defines the distance between two 3D points in the DSM and is given in centimetres. Page 34 of 127

35 Figure 26 Dense matching grid point DSM 3. Orthomosaic : This section allows the user to select the output file format for the orthomosaic as well as different options related to the orthomosaic generation: GeoTIFF (activated by default): Saves the orthomosaic into a GeoTIFF file. The resolution of the orthomosaic is selected in the Local Processing menu of the main window. For more information: For most projects, the orthomosaic is split into several tiles and one GeoTIFF file is generated per tile. o Blending Radius Factor [0,1]: The value can be set between 0 (no blending) and 1 (maximal blending). By default it is set to 0.5. o Use Visibility (activated by default): The DSM is used to determine if a pixel of the orthomosaic is visible from a camera or not. When coloring that pixel / point, only images on which the point is visible are taken into consideration with a specific weight. This option is useful for obtaining orthomosaics with less artefacts when dealing with sharp transition objects such as buildings. o Merge Tiles (activated by default): Generates a single orthomosaic GeoTIFF file containing the whole of the selected area by merging the individual tiles. When this option is deactivated, the merged orthomosaic file is not generated. o GeoTIFF without Transparency: Generates a GeoTIFF file without transparency. Google Maps Tiles and KML: This option allows the user to generate the Google Maps and Google Earth files for the orthomosaic. MapBox Tiles: This option allows the user to generate a Mapbox file for the orthomosaic. 4. Reflectance Map: This section allows the user to save the reflectance map in the following format: GeoTIFF: Saves the reflectance map into a GeoTIFF file. The resolution of the orthomosaic is selected in the Local Processing menu of the main window. For most Page 35 of 127

36 projects, the reflectance map is split into several tiles and one GeoTIFF file is generated per tile. o Merge Tiles (activated by default): Generates a single reflectance map file by merging the individual tiles. When this option is deactivated, the merged reflectance map file is not generated. 6.5 Additional Outputs (Sensefly, 2014) Triangle Model: Allows the user to generate the Triangle Model using the Raster DSM and the orthomosaic while processing step 3. DSM and Orthomosaic Generation. o OBJ Contour Lines: Allows the user to generate the Contour Lines using the Raster DSM while processing step 3. DSM and Orthomosaic Generation. o SHP o PDF o DXF Point Cloud Classification (beta): Allows to set up parameters used when computing the Point Cloud Classification. The Point Cloud Classification classifies the Filtered Point Cloud into two point clouds, the Terrain Point Cloud and the Objects Point Cloud. The Terrain Point Cloud can be used to generate a DTM. 3D PDF: Allows the user to generate the 3D PDF while processing step 3. DSM and Orthomosaic Generation. Page 36 of 127

37 7 Summary of Survey Results The Survey results achieved in the Italian test cases are sufficient to create an accurate DSM & DCM. The DCM can be created only if a DTM already exist. Due to the density of the canopy cover ground points were difficult to determine with a limited view of the sky in the alpine forest. The combined laser scan and UAV data is sufficient to determine species, height and diameter of the trees. To create an accurate forest model I would recommend more scanning be carried rather than sample plots throughout the forest, this could be achieved using new technology such as the Zeb1 handheld scanner rather than the static scanner used in clear cut forest surveying. The combination of data sets from the RGB and NIR cameras helped as the NIR camera lens provided a better view of the forest floor through the canopy. Ground Control Points when sprayed on hard surfaces such as road surface were difficult to recognise with the NIR camera due to their reflective qualities, so further work must be carried out to determine the correct GCP. A new addition to the UAV fleet now provides onboard RTK which will aid accuracy, cut out time wasted setting up ground markers throughout the forest and will enable more ground truthing points to be used to create a DTM. In the following are reported the summary of the surveys in Italy 7.1 Piscine (RGB) Project location Piscine Project date :19 Project size Project georeferencing Area: km 2 / ha / sq. mi. Number of images: 133 Resolution: 9.51 [cm/pixel] 4 GCPs (4 3D), 3.47 m (Mean Sigma Error in X,Y,Z,) GCP name Tolerance XY/Z [m] Error X Error Y [m] Error Z [m] Projection error Verified/Marked 3D GCP: cwl / [ ] [ i l] 8 / 8 3D GCP: cwl / / 13 3D GCP: cwl / / 5 3D GCP: cwl / / 4 Page 37 of 127

38 3D GCP: cwl / / 17 3D GCP: cwl / / 5 3D GCP: cwl / / 6 Mean Sigma RMS error UAV / plane used Camera sensefly ebee Camera maker: Canon Camera model: CanonIXUS127HS Focal length: 4.3 [mm] Image size: 4608x3456 [pixels] Number of flights 1 Flight plan The flight plan was designed taking into account the elevation and slope of this site. Due to the slope of the site, a 3D flight plan was designed in order to obtain a consistent GSD. Flight height: 150 [m] above the ground for each line Frontal overlap ~ 80% Page 38 of 127

39 Lateral overlap ~ 75% Processing settings Version 3.0 Full processing with default settings All available Outputs enabled Processing hardware and timing Desktop computer with: Processor: Intel Core i GHz RAM: 64 GB Hard drive: 7200 RPM HD OS: Windows 7 Graphics: NVIDIA GeForce GTX 680 Desktop Initial rapid processing Initial full processing 20min 01h 02min Results Point cloud Page 39 of 127

40 DSM Orthomosaic Page 40 of 127

41 7.2 Piscine (NIR) Project location Piscine Project date :19 Project size Project georeferencing UAV / plane used Camera Area: km 2 / ha / sq. mi. Number of images: 122 Resolution: [cm/pixel] 3 GCPs (3 3D), m (Mean Sigma Error in X,Y,Z,) GCP name Tolerance Error X [m] Error Y [m] Error Z [m] Projection error Verified/Marked 3D GCP: cwl / / 9 3D GCP: cwl / / 7 3D GCP: cwl / / 2 3D GCP: cwl / #IO 0 / 1 Mean Sigma RMS error (note: red indicates GCP was above tolerance) sensefly ebee Camera maker: Canon Camera model: CanonPowerShotS110 Focal length: 5.2 [mm] Image size: 4000x3000 [pixels] Number of flights 1 Page 41 of 127

42 Flight plan Flight height: 150 [m] above the ground for each line Frontal overlap ~ 80% Lateral overlap ~ 75% Processing settings Version 3.0 Full processing with default settings All available Outputs enabled Processing hardware and timing Desktop computer with: Processor: Intel Core i GHz RAM: 64 GB Hard drive: 7200 RPM HD OS: Windows 7 Graphics: NVIDIA GeForce GTX 680 Desktop Initial rapid processing Initial full processing 20min 01h 08min Page 42 of 127

43 Results Point cloud DSM Page 43 of 127

44 Orthomosaic 7.3 Montesover (RGB) Project location Montesover Project date :16 Project size Project georeferencing Area: km 2 / ha / sq. mi. Number of images: 89 Resolution: [cm/pixel] 4 GCPs (4 3D), 7.03 m (Mean Sigma Error in X,Y,Z,) GCP name Tolerance Error X [m] Error Y [m] Error Z [m] Projection Verified/Marked 3D GCP: cwl / / 10 3D GCP: cwl / / 15 3D GCP: cwl / / 12 3D GCP: cwl / / 11 Page 44 of 127

45 3D GCP: cwl / #IO 0 / 2 3D GCP: cwl / / 12 3D GCP: cwl / / 15 Mean Sigma RMS error UAV / plane used Camera sensefly ebee Camera maker: Canon Camera model: CanonIXUS127HS Focal length: 4.3 [mm] Image size: 4608x3456 [pixels] Number of flights 1 Flight plan Flight height: 150 [m] Frontal overlap ~ 80% Lateral overlap ~ 75% Processing settings Version 3.0 Full processing with default settings Page 45 of 127

46 All available Outputs enabled Processing hardware and timing Desktop computer with: Processor: Intel Core i GHz RAM: 64 GB Hard drive: 7200 RPM HD OS: Windows 7 Graphics: NVIDIA GeForce GTX 680 Desktop Initial rapid processing Initial full processing 22min 01h 10min Results Point cloud Page 46 of 127

47 SLOPE - Integrated processing and control systems for sustainable forest Production in DSM Orthomosaic Page 47 of 127

48 7.4 Montesover (NIR) Project location Montesover Project date :40 Project size Project georeferencing UAV / plane used Camera Area: km 2 / ha / sq. mi. Number of images: 309 Resolution: [cm/pixel] 5 GCPs (5 3D), 5.55 m (Mean Sigma Error in X,Y,Z,) GCP name Tolerance Error X [m] Error Y [m] Error Z [m] Projection Verified/Marked 3D GCP: cwl / / 13 3D GCP: cwl / / 4 3D GCP: cwl / / 11 3D GCP: cwl / / 9 3D GCP: cwl / / 9 3D GCP: cwl / / 17 Mean Sigma RMS error sensefly ebee Camera maker: Canon Camera model: CanonPowerShotS110 Focal length: 5.2 [mm] Image size: 4000x3000 [pixels] Number of flights 1 Page 48 of 127

49 Flight plan Flight height: 150 [m] above the ground for each line Frontal overlap ~ 80% Lateral overlap ~ 75% Processing settings Version 3.0 Full processing with default settings All available Outputs enabled Processing hardware and timing Desktop computer with: Processor: Intel Core i GHz RAM: 64 GB Hard drive: 7200 RPM HD OS: Windows 7 Graphics: NVIDIA GeForce GTX 680 Desktop Initial rapid processing Initial full processing 20min 02h 29min Results Page 49 of 127

50 SLOPE - Integrated processing and control systems for sustainable forest Production in Point cloud DSM Page 50 of 127

51 Orthomosaic Page 51 of 127

52 8 Annex 8.1 Quality Report Outputs Below are the auto-generated Quality Reports from Postflightterra3D Generated with version Important: Click on the different icons for: Help to analyze the results in the Quality Report Additional information about the feature For additional tips to analyze the Quality Report, click here. Site 1 Montesover Summary Page 52 of 127

53 Project fp7-f3-f2-area2-montesova-nir+rgb Processed 2014-Nov-04 10:57:43 Camera Model Name CanonPowerShotS110_5.2_4000x3000 (NIR,Green,Blue) Camera Model Name CanonIXUS127HS_4.3_4608x3456 (RGB) Average Ground Sampling Distance (GSD) cm / 5.28 in Area Covered km 2 / ha / sq. mi. / acres Image Coordinate System WGS84 Ground Control Point (GCP) Coordinate System WGS 84 / UTM zone 32N Output Coordinate System WGS84 / UTM zone 32N Processing Type full aerial nadir Feature Extraction Image Scale 1 Camera Model Parameter Optimization optimize externals and all internals Time for Initial Processing (without report) 01h:04m:05s Quality Check Page 53 of 127

54 Images median of keypoints per image Dataset 395 out of 398 images calibrated (99%), all images enabled Camera Optimization 0.45% relative difference between initial and final focal length Matching median of matches per calibrated image Georeferencing 7 GCPs (7 3D), mean error = m Preview Figure 1: Orthomosaic and the corresponding sparse Digital Surface Model (DSM) before densification. Page 54 of 127

55 Calibration Details Number of Calibrated Images 395 out of 398 Number of Geolocated Images 398 out of 398 Initial Image Positions Figure 2: Top view of the initial image position. The green line follows the position of the images in time starting from the large blue dot. Page 55 of 127

56 Computed Image/GCPs/Manual Tie Points Positions Figure 3: Offset between initial (blue dots) and computed (green dots) image positions as well as the offset between the GCPs initial positions (blue crosses) and their computed positions (green crosses) in the top-view (XY plane), front-view (XZ plane), and side-view (YZ plane). Red dots indicate disabled or uncalibrated images. Page 56 of 127

57 Overlap Number of overlapping images: Figure 4: Number of overlapping images computed for each pixel of the orthomosaic. Red and yellow areas indicate low overlap for which poor results may be generated. Green areas indicate an overlap of over 5 images for every pixel. Good quality results will be generated as long as the number of keypoint matches is also sufficient for these areas (see Figure 5 for keypoint matches). Bundle Block Adjustment Details Number of 2D Keypoint Observations for Bundle Block Adjustment Number of 3D Points for Bundle Block Adjustment Mean Reprojection Error [pixels] Page 57 of 127

58 Internal Camera Parameters CanonPowerShotS110_5.2_4000x3000 (NIR,Green,Blue). Sensor Dimensions: 7.44 [mm] x 5.58 [mm] EXIF ID: CanonPowerShotS110_5.2_4000x3000 Focal Length Principal Point x Principal Point y R1 R2 R3 T1 T2 Initial Values [pixel] [mm] [pixel] [mm] [pixel] [mm] Optimized Values [pixel] [mm] [pixel] [mm] [pixel] [mm] Internal Camera Parameters CanonIXUS127HS_4.3_4608x3456 (RGB). Sensor Dimensions: [mm] x [mm] EXIF ID: CanonIXUS127HS_4.3_4608x3456 Focal Length Principal Point x Principal Point y R1 R2 R3 T1 T2 Page 58 of 127

59 Initial Values [pixel] [mm] [pixel] [mm] [pixel] [mm] Optimized Values [pixel] [mm] [pixel] [mm] [pixel] [mm] D Keypoints Table Number of 2D Keypoints per Image Number of Matched 2D Keypoints per Image Median Min Max Mean D Keypoints Table for Camera CanonPowerShotS110_5.2_4000x3000 (NIR,Green,Blue) Number of 2D Keypoints per Image Number of Matched 2D Keypoints per Image Page 59 of 127

60 Median Min Max Mean D Keypoints Table for Camera CanonIXUS127HS_4.3_4608x3456 (RGB) Number of 2D Keypoints per Image Number of Matched 2D Keypoints per Image Median Min Max Mean Median / 75% / Maximal Number of Matches Between Camera Models CanonPowerShotS110_5.2_4000x 3000 (NIR,Green,Blue) CanonIXUS127HS_4.3_4608x3 456 (RGB) CanonPowerShotS110_5.2_4000x 3000 (NIR,Green,Blue) 26 / 177 / 8440 CanonIXUS127HS_4.3_4608x3456 (RGB) 14 / 84 / Page 60 of 127

61 3D Points from 2D Keypoint Matches Number of 3D Points Observed In 2 Images In 3 Images In 4 Images In 5 Images Page 61 of 127

62 In 6 Images In 7 Images In 8 Images 9565 In 9 Images 6384 In 10 Images 4368 In 11 Images 3193 In 12 Images 2428 In 13 Images 1734 In 14 Images 1367 In 15 Images 1033 In 16 Images 922 In 17 Images 748 In 18 Images 578 In 19 Images 503 In 20 Images 421 Page 62 of 127

63 In 21 Images 328 In 22 Images 285 In 23 Images 219 In 24 Images 231 In 25 Images 202 In 26 Images 159 In 27 Images 169 In 28 Images 104 In 29 Images 99 In 30 Images 76 In 31 Images 81 In 32 Images 70 In 33 Images 66 In 34 Images 62 In 35 Images 56 Page 63 of 127

64 In 36 Images 44 In 37 Images 37 In 38 Images 37 In 39 Images 32 In 40 Images 36 In 41 Images 22 In 42 Images 20 In 43 Images 22 In 44 Images 25 In 45 Images 24 In 46 Images 12 In 47 Images 12 In 48 Images 20 In 49 Images 19 In 50 Images 12 Page 64 of 127

65 In 51 Images 21 In 52 Images 7 In 53 Images 16 In 54 Images 9 In 55 Images 11 In 56 Images 8 In 57 Images 15 In 58 Images 8 In 59 Images 3 In 60 Images 8 In 61 Images 1 In 62 Images 6 In 63 Images 3 In 64 Images 5 In 65 Images 4 Page 65 of 127

66 In 66 Images 1 In 67 Images 1 In 68 Images 5 In 69 Images 3 In 70 Images 4 In 71 Images 1 In 72 Images 3 In 73 Images 2 In 74 Images 5 In 75 Images 2 In 76 Images 2 In 77 Images 2 In 78 Images 1 In 79 Images 1 In 81 Images 1 Page 66 of 127

67 In 82 Images 1 In 83 Images 2 In 84 Images 1 In 88 Images 1 In 89 Images 1 In 97 Images 2 In 104 Images 1 In 105 Images 1 In 107 Images 1 In 115 Images 1 In 116 Images 1 In 117 Images 1 In 129 Images 1 Page 67 of 127

68 3D Points from 2D Keypoint Matches Number of matches Figure 5: Top view of the image computed positions with a link between matching images. The darkness of the links indicates the number of matched 2D keypoints between the images. Bright links indicate weak links and require manual tie points or more images. Geolocation Details Page 68 of 127

69 Ground Control Points GCP Name Accuracy XY/Z [m] Error X [m] Error Y [m] Error Z [m] Projection Error [pixel] Verified/Marked cwl1 (3D) 0.020/ / 23 cwl2 (3D) 0.020/ / 19 cwl3 (3D) 0.020/ / 23 cwl4 (3D) 0.020/ / 20 cwl5 (3D) 0.020/ / 2 cwl7 (3D) 0.020/ / 21 cwl6 (3D) 0.020/ / 32 Mean Sigma RMS Error Localisation accuracy per GCP and mean errors in the three coordinate directions. The last column counts the number of images where the GCP has been automatically verified vs. manually marked. Page 69 of 127

70 Absolute Geolocation Variance Min Error [m] Max Error [m] Geolocation Error X [%] Geolocation Error Y [%] Geolocation Error Z [%] Page 70 of 127

71 Mean Sigma RMS Error Min Error and Max Error represent geolocation error intervals between -1.5 and 1.5 times the maximum accuracy of all the images. Columns X, Y, Z show the percentage of images with geolocation errors within the predefined error intervals. The geolocation error is the difference between the initial and computed image positions. Note that the image geolocation errors do not correspond to the accuracy of the observed 3D points. Page 71 of 127

72 Geotag Orientational Variance RMS [degree] Omega Phi Kappa Geolocation RMS error of the orientation angles given by the difference between the initial and computed image orientation angles. Georeference Verification Page 72 of 127

73 GCP Name: cwl1 ( , ,854.40) IMG_3294.JPG IMG_3295.JPG IMG_3296.JPG IMG_3297.JPG IMG_3300.JPG IMG_3301.JPG IMG_3302.JPG IMG_3305.JPG IMG_3306.JPG IMG_3307.JPG IMG_0583.JPG IMG_0584.JPG IMG_0589.JPG IMG_0590.JPG IMG_0596.JPG IMG_0597.JPG IMG_0603.JPG IMG_0604.JPG IMG_0650.JPG IMG_0651.JPG IMG_0691.JPG IMG_0692.JPG IMG_0695.JPG Page 73 of 127

74 Page 74 of 127

75 GCP cwl1 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_3286.JPG IMG_3287.JPG IMG_3288.JPG IMG_3289.JPG IMG_3290.JPG IMG_3291.JPG Page 75 of 127

76 GCP Name: cwl2 ( , ,861.45) IMG_3287.JPG IMG_3288.JPG IMG_3289.JPG IMG_3290.JPG IMG_3291.JPG IMG_3292.JPG IMG_3296.JPG IMG_3297.JPG IMG_3298.JPG IMG_3299.JPG IMG_3300.JPG IMG_3301.JPG IMG_3306.JPG IMG_3307.JPG IMG_3308.JPG IMG_0595.JPG IMG_0596.JPG IMG_0650.JPG IMG_0651.JPG Page 76 of 127

77 GCP cwl2 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_3286.JPGIMG_3293.JP G IMG_3294.JPG IMG_3295.JPG IMG_3302.JPG IMG_3303.JPG Page 77 of 127

78 GCP Name: cwl3 ( , , ) IMG_3354.JPG IMG_3356.JPG IMG_3357.JPG IMG_3360.JPG IMG_3361.JPG IMG_3362.JPG IMG_3363.JPG IMG_3366.JPG IMG_3367.JPG IMG_3368.JPG IMG_3371.JPG IMG_3372.JPG IMG_0452.JPG IMG_0453.JPG IMG_0455.JPG IMG_0456.JPG IMG_0457.JPG IMG_0458.JPG IMG_0671.JPG IMG_0672.JPG IMG_0673.JPG IMG_0714.JPG IMG_0715.JPG Page 78 of 127

79 Page 79 of 127

80 GCP cwl3 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_3365.JPG IMG_3370.JPG IMG_3373.JPG IMG_0451.JPG IMG_0461.JPG IMG_0462.JPG Page 80 of 127

81 GCP Name: cwl4 ( , , ) IMG_3357.JPG IMG_3361.JPG IMG_3362.JPG IMG_3363.JPG IMG_3365.JPG IMG_3366.JPG IMG_3367.JPG IMG_3371.JPG IMG_3372.JPG IMG_3373.JPG IMG_3374.JPG IMG_0451.JPG IMG_0452.JPG IMG_0456.JPG IMG_0457.JPG IMG_0458.JPG IMG_0462.JPG IMG_0673.JPG IMG_0714.JPG IMG_0715.JPG GCP cwl4 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_0757.JPG Page 81 of 127

82 GCP Name: cwl5 ( , , ) IMG_3361.JPG IMG_3362.JPG GCP cwl5 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_3357.JPG IMG_3360.JPG IMG_3363.JPG IMG_3365.JPG IMG_3366.JPG IMG_3367.JPG Page 82 of 127

83 GCP Name: cwl7 ( , , ) IMG_3347.JPG IMG_3348.JPG IMG_3349.JPG IMG_3353.JPG IMG_3354.JPG IMG_3355.JPG IMG_3356.JPG IMG_3357.JPG IMG_3358.JPG IMG_3362.JPG IMG_3363.JPG IMG_3364.JPG IMG_0458.JPG IMG_0461.JPG IMG_0462.JPG IMG_0471.JPG IMG_0472.JPG IMG_0473.JPG IMG_0713.JPG IMG_0717.JPG IMG_0718.JPG GCP cwl7 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_3352.JPG IMG_3360.JPG IMG_3361.JPG IMG_3365.JPG Page 83 of 127

84 IMG_3366.JPG IMG_3367.JPG Page 84 of 127

85 GCP Name: cwl6 ( , , ) IMG_3348.JPG IMG_3353.JPG IMG_3354.JPG IMG_3356.JPG IMG_3357.JPG IMG_3362.JPG IMG_3363.JPG IMG_3364.JPG IMG_3365.JPG IMG_3366.JPG IMG_3367.JPG IMG_3371.JPG IMG_3372.JPG IMG_3373.JPG IMG_3374.JPG IMG_0451.JPG IMG_0457.JPG IMG_0458.JPG IMG_0459.JPG IMG_0461.JPG IMG_0462.JPG IMG_0463.JPG IMG_0472.JPG IMG_0673.JPG IMG_0674.JPG IMG_0675.JPG IMG_0713.JPG IMG_0714.JPG IMG_0715.JPG IMG_0716.JPG IMG_0717.JPG IMG_0757.JPG Page 85 of 127

86 SLOPE - Integrated processing and control systems for sustainable forest Production in Page 86 of 127

87 GCP cwl6 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_3361.JPG IMG_0450.JPG IMG_0452.JPG IMG_0456.JPG IMG_0470.JPG IMG_0471.JPG Figure 7: Images in which GCPs have been marked (yellow circle) and in which their computed 3D points have been projected (green circle). A green circle outside of the yellow circle indicates either an accuracy issue or a GCP issue. Point Cloud Densification details Summary Page 87 of 127

88 Processing Type aerial nadir Image Scale multiscale, 1/2 (half image size, default) Point Density optimal Minimum Number of Matches 3 Use Densification Area yes Use Annotations yes Use Noise Filtering yes, radius = 10 GSD Use Surface Smoothing yes, sharp, radius = 10 GSD Time for Densification and Filtering (without report) 02h:49m:45s Results Number of Processed Clusters 2 Number of 3D Densified Points Number of 3D Filtered Points Average Density (per m 3 ) 1.91 Page 88 of 127

89 Area 2 Piscine Summary Project fp7-f4-area1-piscine-nir+rgb Processed 2014-Nov-03 16:16:23 Camera Model Name CanonIXUS127HS_4.3_4608x3456 (RGB) Camera Model Name CanonPowerShotS110_5.2_4000x3000 (NIR,Green,Blue) Average Ground Sampling Distance (GSD) 7.64 cm / 3 in Area Covered km 2 / ha / sq. mi. / acres Image Coordinate System WGS84 Ground Control Point (GCP) Coordinate System WGS 84 / UTM zone 32N Page 89 of 127

90 Output Coordinate System WGS 84 / UTM zone 32N Processing Type full aerial nadir Feature Extraction Image Scale 1 Camera Model Parameter Optimization optimize externals and all internals Time for Initial Processing (without report) 52m:23s Quality Check Images median of keypoints per image Dataset 253 out of 255 images calibrated (99%), all images enabled Camera Optimization 0.33% relative difference between initial and final focal length Matching median of matches per calibrated image Georeferencing 7 GCPs (7 3D), mean error = m Page 90 of 127

91 Preview Figure 1: Orthomosaic and the corresponding sparse Digital Surface Model (DSM) before densification. Calibration Details Number of Calibrated Images 253 out of 255 Number of Geolocated Images 255 out of 255 Page 91 of 127

92 Initial Image Positions Figure 2: Top view of the initial image position. The green line follows the position of the images in time starting from the large blue dot. Computed Image/GCPs/Manual Tie Points Positions Page 92 of 127

93 Figure 3: Offset between initial (blue dots) and computed (green dots) image positions as well as the offset between the GCPs initial positions (blue crosses) and their computed positions (green crosses) in the top-view (XY plane), front-view (XZ plane), and side-view (YZ plane). Red dots indicate disabled or uncalibrated images. Page 93 of 127

94 Overlap Number of overlapping images: Figure 4: Number of overlapping images computed for each pixel of the orthomosaic. Red and yellow areas indicate low overlap for which poor results may be generated. Green areas indicate an overlap of over 5 images for every pixel. Good quality results will be generated as long as the number of keypoint matches is also sufficient for these areas (see Figure 5 for keypoint matches). Bundle Block Adjustment Details Page 94 of 127

95 Number of 2D Keypoint Observations for Bundle Block Adjustment Number of 3D Points for Bundle Block Adjustment Mean Reprojection Error [pixels] Internal Camera Parameters CanonIXUS127HS_4.3_4608x3456 (RGB). Sensor Dimensions: 6.17 [mm] x 4.63 [mm] EXIF ID: CanonIXUS127HS_4.3_4608x3456 Focal Length Principal Point x Principal Point y R1 R2 R3 T1 T2 Initial Values [pixel] [mm] [pixel] [mm] [pixel] [mm] Optimized Values [pixel] [mm] [pixel] [mm] [pixel] [mm] Page 95 of 127

96 Internal Camera Parameters CanonPowerShotS110_5.2_4000x3000 (NIR,Green,Blue). Sensor Dimensions: [mm] x [mm] EXIF ID: CanonPowerShotS110_5.2_4000x3000 Focal Length Principal Point x Principal Point y R1 R2 R3 T1 T2 Initial Values [pixel] [mm] [pixel] [mm] [pixel] [mm] Optimized Values [pixel] [mm] [pixel] [mm] [pixel] [mm] D Keypoints Table Number of 2D Keypoints per Image Number of Matched 2D Keypoints per Image Page 96 of 127

97 Median Min Max Mean D Keypoints Table for Camera CanonIXUS127HS_4.3_4608x3456 (RGB) Number of 2D Keypoints per Image Number of Matched 2D Keypoints per Image Median Min Max Mean D Keypoints Table for Camera CanonPowerShotS110_5.2_4000x3000 (NIR,Green,Blue) Number of 2D Keypoints per Image Number of Matched 2D Keypoints per Image Median Min Page 97 of 127

98 Max Mean Median / 75% / Maximal Number of Matches Between Camera Models CanonIXUS127HS_4.3_4608x3 456 (RGB) CanonPowerShotS110_5.2_4000x 3000 (NIR,Green,Blue) CanonIXUS127HS_4.3_4608x3456 (RGB) 14 / 85 / 7195 CanonPowerShotS110_5.2_4000x 3000 (NIR,Green,Blue) 10 / 58 / D Points from 2D Keypoint Matches Number of 3D Points Observed Page 98 of 127

99 In 2 Images In 3 Images In 4 Images In 5 Images 7742 In 6 Images 3507 In 7 Images 1730 In 8 Images 1013 In 9 Images 599 In 10 Images 385 In 11 Images 278 In 12 Images 170 In 13 Images 129 In 14 Images 65 In 15 Images 63 In 16 Images 46 Page 99 of 127

100 In 17 Images 38 In 18 Images 17 In 19 Images 12 In 20 Images 5 In 21 Images 8 In 22 Images 8 In 23 Images 5 In 24 Images 7 In 25 Images 4 In 26 Images 6 In 27 Images 4 In 28 Images 4 In 29 Images 1 In 30 Images 1 In 31 Images 5 Page 100 of 127

101 In 33 Images 1 In 37 Images 1 In 38 Images 1 In 39 Images 1 3D Points from 2D Keypoint Matches Number of matches Figure 5: Top view of the image computed positions with a link between matching images. The darkness of the links indicates the number of matched 2D keypoints between the images. Bright links indicate weak links and require manual tie points or more images. Page 101 of 127

102 Geolocation Details Ground Control Points GCP Name Accuracy XY/Z [m] Error X [m] Error Y [m] Error Z [m] Projection Error [pixel] Verified/Marked cwl1 (3D) 0.020/ / 8 cwl3 (3D) 0.020/ / 20 cwl4 (3D) 0.020/ / 5 cwl8 (3D) 0.020/ / 10 cwl5 (3D) 0.020/ / 18 cwl6 (3D) 0.020/ / 12 Page 102 of 127

103 cwl7 (3D) 0.020/ / 13 Mean Sigma RMS Error Localisation accuracy per GCP and mean errors in the three coordinate directions. The last column counts the number of images where the GCP has been automatically verified vs. manually marked Absolute Geolocation Variance Page 103 of 127

104 Min Error [m] Max Error [m] Geolocation Error X [%] Geolocation Error Y [%] Geolocation Error Z [%] Mean Sigma Page 104 of 127

105 RMS Error Min Error and Max Error represent geolocation error intervals between -1.5 and 1.5 times the maximum accuracy of all the images. Columns X, Y, Z show the percentage of images with geolocation errors within the predefined error intervals. The geolocation error is the difference between the intial and computed image positions. Note that the image geolocation errors do not correspond to the accuracy of the observed 3D points. Geotag Orientational Variance RMS [degree] Omega Phi Kappa Geolocation RMS error of the orientation angles given by the difference between the initial and computed image orientation angles. Georeference Verification Page 105 of 127

106 GCP Name: cwl1 ( , ,872.34) IMG_2848.JPG IMG_2862.JPG IMG_2863.JPG IMG_2865.JPG IMG_2866.JPG IMG_2878.JPG IMG_2879.JPG IMG_2880.JPG GCP cwl1 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_0338.JPG IMG_0339.JPG IMG_0351.JPG IMG_0352.JPG IMG_0353.JPG IMG_0354.JPG Page 106 of 127

107 Page 107 of 127

108 GCP Name: cwl3 ( , ,870.54) IMG_0339.JPG IMG_0341.JPG IMG_0347.JPG IMG_0348.JPG IMG_0349.JPG IMG_0350.JPG IMG_0351.JPG IMG_2867.JPG IMG_2868.JPG IMG_2869.JPG IMG_2874.JPG IMG_2875.JPG IMG_2876.JPG IMG_2877.JPG IMG_2883.JPG IMG_2884.JPG IMG_2885.JPG IMG_2891.JPG IMG_2892.JPG IMG_2893.JPG GCP cwl3 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_0340.JPG IMG_0356.JPG IMG_0357.JPG Page 108 of 127

109 IMG_0358.JPG IMG_0359.JPG IMG_0363.JPG Page 109 of 127

110 GCP Name: cwl4 ( , ,916.74) IMG_2860.JPG IMG_2861.JPG IMG_2918.JPG IMG_2925.JPG IMG_2934.JPG GCP cwl4 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_0348.JPG IMG_0349.JPG IMG_0350.JPG IMG_0351.JPG IMG_0356.JPG IMG_0357.JPG Page 110 of 127

111 GCP Name: cwl8 ( , ,984.96) IMG_0364.JPG IMG_0376.JPG IMG_0381.JPG IMG_0425.JPG IMG_2835.JPG IMG_2841.JPG IMG_2935.JPG IMG_2936.JPG IMG_2939.JPG IMG_2940.JPG GCP cwl8 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_0365.JPG IMG_0366.JPG IMG_0374.JPG IMG_0375.JPG IMG_0380.JPG Page 111 of 127

112 IMG_0382.JPG Page 112 of 127

113 GCP Name: cwl9 ( , ,922.12) IMG_0358.JPG GCP cwl9 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_0339.JPG IMG_0347.JPG IMG_0348.JPG IMG_0349.JPG IMG_0350.JPG IMG_0356.JPG Page 113 of 127

114 GCP Name: cwl5 ( , ,938.01) IMG_2849.JPG IMG_2850.JPG IMG_2851.JPG IMG_2859.JPG IMG_2861.JPG IMG_2900.JPG IMG_2901.JPG IMG_2902.JPG IMG_2908.JPG IMG_2909.JPG IMG_2910.JPG IMG_2918.JPG IMG_2919.JPG IMG_2924.JPG IMG_2934.JPG IMG_2940.JPG IMG_2950.JPG IMG_2952.JPG Page 114 of 127

115 SLOPE - Integrated processing and control systems for sustainable forest Production in Page 115 of 127

116 GCP cwl5 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_0357.JPG IMG_0358.JPG IMG_0364.JPG IMG_0365.JPG IMG_0366.JPG IMG_0367.JPG Page 116 of 127

117 GCP Name: cwl6 ( , ,997.63) IMG_0364.JPG IMG_0376.JPG IMG_0425.JPG IMG_2835.JPG IMG_2841.JPG IMG_2853.JPG IMG_2936.JPG IMG_2938.JPG IMG_2939.JPG IMG_2940.JPG IMG_2951.JPG IMG_2953.JPG GCP cwl6 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_0365.JPG IMG_0366.JPG IMG_0374.JPG IMG_0375.JPG IMG_0380.JPG IMG_0381.JPG Page 117 of 127

118 Page 118 of 127

119 GCP Name: cwl7 ( , ,952.48) IMG_0366.JP G IMG_0376.JP G IMG_2835.JP G IMG_2839.JP G IMG_2840.JP G IMG_2841.JP G IMG_2852.JP G IMG_2925.JP G IMG_2934.JP G IMG_2935.JP G IMG_2938.JP G IMG_2939.JP G IMG_2952.JP G Page 119 of 127

120 Page 120 of 127

121 GCP cwl7 was not marked in the following images (only up to 6 images shown). If the circle is too far away from the initial GCP position, also measure the GCP in these images to improve the accuracy. IMG_0357.JP G IMG_0358.JP G IMG_0359.JP G IMG_0363.JP G IMG_0364.JP G IMG_0365.JP G Figure 7: Images in which GCPs have been marked (yellow circle) and in which their computed 3D points have been projected (green circle). A green circle outside of the yellow circle indicates either an accuracy issue or a GCP issue. Point Cloud Densification details Summary Page 121 of 127