DOI: http://doi.org/10.3846/enviro.2017.XXX
STUDENT CONTEST 2018
Accuracy of Stockpile Volume Calculations Based on UAV Photogrammetry Kaupo Kokamägi1
1
Institute of Forestry and Rural Engineering, Estonian University of Life Sciences, Tartu, Estonia 1 REIB OÜ, Tartu, Estonia E-mail: kaupo.kokamagi@gmail.com
Abstract. Due to the overall development of technology, UAVs (Unmanned Aerial Vehicles) are being used mode widely. Unmanned aerial vehicles and the cameras have been developing fast during the recent years. The popularity of UAVs has caused mass production and lower prices of these instruments. Generally the fast development of both hardware and software has made it possible to use unmanned aerial vehicles in geodesy. This article is based on a masters thesis, published in Estonian University of Life Sciences, 2018 (Kokamägi, 2018). The aim of current study was to assess the compliance of accuracy of stockpile volume calculations based on UAV photogrammetry to current laws in Estonia. Two different UAVs and two different objects were compared in this study. The accuracy of photogrammetric models was also compared to the accuracy of a model based on RTK GNSS survey. Amount of time spent on different parts of the project was also evaluated. Data used in current study was collected in autumn of 2017 in Laiküla peat extraction area in Lääne County, Estonia and spring of 2018 in Karude gravel pit in Järva County, Estonia. Observed objects were a regularly shaped peat stockpile and an irregularly shaped gravel stockpile. On the first site, data was collected with a terrestrial laser scanner, a GNSS device and two different UAVs. On the second site data was collected with a terrestrial laser scanner, a GNSS device and one UAV. For accuracy assessment, volume of different models was compared to the volume of the models based on laser scanning data. The study found that the RMSE (Root Mean Square Error) of different photogrammetrical models of the Laiküla object (722,52 m3) was 14,31 m3 and the RMSE of different photogrammetrical models of the Karude object (674,04 m3) was 20,95 m3. Relative errors of all of the photogrammetrical models were under 4%, which is smaller than the allowed error of 12% in Estonia. Relative error of RTK GNSS model of Laiküla object was 5,81% and relative errors of different photogrammetric models of the same object were between 0,70% and 3,19%. Respective errors of Karude object were 3,28% and 2,32% to 3,70%. It was found that the advantages of UAV photogrammetry become apparent as the size and complexity of the objects grow. Results show that using UAV photogrammetry to determine stockpile volumes is sufficiently accurate with both of the tested UAVs. Still, it is important that using UAVs for geodetic work would be done by trained professionals and all the legal requirements would be met. It was found that sufficient accuracy was also achieved without using the GCPs (Ground Control Points). However, using GCPs did increase the accuracy. Furthermore, using different types of GCPs had no impact to the accuracy of the volume. Still, using more GCPs also increased the accuracy.
Introduction Due to the overall development of technology, UAVs (Unmanned Aerial Vehicles) are being used mode widely. It is mainly used in dangerous areas or areas with difficult accessibility and also when measuring large objects, where the accuracy of mapping doesn’t have to be very high. Unmanned aerial vehicles and the cameras have been developing fast during the recent years. The popularity of UAVs has caused mass production and lower prices of these instruments. Development of photogrammetry software has made it possible to use commercial cameras for photogrammetrical purposes. Generally the fast development of both hardware and software has made it possible to use unmanned aerial vehicles in geodesy. The aim of current study was to assess the compliance of accuracy of stockpile volume calculations based on UAV photogrammetry to current laws in Estonia. This is important because land surveyors have an interest in finding effective ways to survey and assessing the accuracy of these ways. Furthermore it was investigated how using different GCP’s and different characteristics of different objects affect the results of the survey.
Address Offices in Brussels : Rue du Nord 76, BE – 1000 Bruxelles. Tel +32/2/217.39.72 Fax +32/2/219.31.47 E-mail: maurice.barbieri@clge.eu - www.clge.eu EU-Transparency Register of interest representatives - 510083513941-24
STUDENT CONTEST 2018 Two different UAVs and two different objects were compared in this study. The accuracy of photogrammetric models was also compared to the accuracy of a model based on RTK GNSS survey. Amount of time spent on different parts of the project was also evaluated. Data used in current study was collected in autumn of 2017 in Laiküla peat extraction area in Lääne County, Estonia and spring of 2018 in Karude gravel pit in Järva County, Estonia. The observed objects were a regularly shaped peat stockpile and an irregularly gravel stockpile. On the first site, data was collected with a terrestrial laser scanner, a GNSS device and two different UAVs. On the second site data was collected with a terrestrial laser scanner, a GNSS device and one UAV. For accuracy assessment, volume of different models was compared to the volume of the models based on laser scanning data. Following hypotheses were set for this study: Accuracy of volume calculations based on UAV photography does not exceed the allowed 12% difference in Estonia (Markšeideritöö kord 2012, § 4 lg 3). Using aerosol paint does not affect the accuracy of results compared to using special GCP’s. Using UAV’s for volume surveys reduces the time spent on fieldwork significally. Materials and methods Two different object were chosen for this study. The first object was a regularly shaped peat stockpile in Laiküla peat extraction area in Lääne County, Estonia (Fig. 1). The second object was an irregularly shaped gravel stockpile in Karude gravel pit in Järva County, Estonia (Fig. 1).
Fig. 1. The location of Laiküla peat extraction area marked red and Karude gravel pit marked blue (Web mapping service of Estonian Land Board)
The area of the peat stockpile was 463 m2 and the area of the gravel stockpile was 394 m2. Both Laiküla peat extraction area and Karude gravel pit are in active use and they are both typical objects that require regular volume surveys. Trimble R4-3 GNSS device was used for surveying GCP’s and the contours of both stockpiles. Both of the objects were scanned with Trimble SX10 scanning total station. DJI Phantom 4 pro v2.0 UAV was used for surveying both of the objects (Fig. 2). It is a low-priced and widely used commercial UAV which has a 20 MP (megapixel) integrated camera. Aibotix Aibot X6 was used only in Laiküla peat extraction area (Fig. 2). It is an UAV, which is made specifically for photogrammetric use. It was carrying a 42.4 MP Sony ILCE-7RM2 camera during this study.
2
Address Offices in Brussels : Rue du Nord 76, BE – 1000 Bruxelles. Tel +32/2/217.39.72 Fax +32/2/219.31.47 E-mail: maurice.barbieri@clge.eu - www.clge.eu EU-Transparency Register of interest representatives - 510083513941-24
STUDENT CONTEST 2018
Fig. 2. DJI Phantom 4 Pro v2.0 on the left and Aibotix Aibot X6 on the right in Laiküla peat extraction area. (Photo: Mart Rae) Flight planning for Aibot X6 was done with AiProFlight software and for DJI Phantom 4 pro with DroneDeploy software. The point clouds were processed with Trimble Business Center and Autodesk Recap 2019 software. Images were oriented and point clouds were generated with Agisoft Photoscan Professional 1.4.0 software. Drawing software Autodesk Civil 2019 was used for creating models from point clouds and calculating the volumes. Microsoft Excel was used for analysis of the results. The first step of work process was choosing the objects and instruments. Fieldwork started with object preparation, which included setting up the GCP’s and measuring them with a RTK GNSS device. After that, the object was surveyed with a RTK GNSS device and scanned with a terrestrial laser scanner. Flight planning and photogrammetric flight were conducted after scanning. After collecting data, a quick data check was done on site and then the GCP’s were collected. That concludes the fieldwork. After that data was processed, which included creating models of the objects from different data sources and calculating the volumes. Then the volumes of different models were compared to each other and both relative and absolute difference from the scan based models were found. Collecting data in Laiküla peat extraction area Surveying peat stockpiles manually is often difficult and dangerous. The stockpile had a regular shape which made surveying the bottom contour easier (Fig. 3). Peat stockpiles need to be surveyed several times per year and it would be useful to make this job more effective. Surveying took place on 24th of October 2017. Thanks to the cold weather, it was possible to manually measure the ridge of the stockpile, because one slope of it was frozen. During object preparation 12 aerosol painted GCP’s and 9 special photogrammetric GCP’s were set up around and on top of the stockpile. After this the GCP’s coordinates and the object were surveyed with RTK GNSS device. The stockpile was scanned from four locations with Trimble SX10 scanning total station. This object was surveyed with both Aibotix Aibot X6 and DJI Phantom 4 Pro. DJI flight was planned with DroneDeploy software, which calculated the optimal trajectory. DJI flight took under 5 minutes. 415 photos were taken, of which we used 76, which were taken on the right height and included the object or GCP’s. The height of the flight was 33 metres and ground pixel size was 8 millimetres.
Address Offices in Brussels : Rue du Nord 76, BE – 1000 Bruxelles. Tel +32/2/217.39.72 Fax +32/2/219.31.47 E-mail: maurice.barbieri@clge.eu - www.clge.eu EU-Transparency Register of interest representatives - 510083513941-24
3
STUDENT CONTEST 2018
Fig. 3. The selected object in Laiküla peat extraction area is on the right.
Aibot X6 flight was planned with AiProFlight software, the planning and setting up the UAV took about 15 minutes, which is longer than with the DJI Phantom. The flight took about 5 minutes. 95 images were taken, of which we used 48. The height of the flight was 47 metres and ground pixel size was 6 millimetres. Collecting data in Karude gravel pit The other object was gravel stockpile (Fig. 4), which is better to survey manually, but open gravel pits are another type of objects that can potentially be surveyed a lot faster with the help of UAV’s. The chosen object had an irregular shape, which made manual surveying more difficult and allowed us to compare the results with the regularly shaped peat stockpile survey. Surveying took place on 10th of April 2018. During object preparation 18 special photogrammetric GCP’s were set up around and on top of the stockpile. After this the GCP’s coordinates and the object were surveyed with RTK GNSS device. The stockpile was scanned from eight locations with Trimble SX10 scanning total station.
Fig. 4. The selected object in Karude gravel pit. Setting up GCP’s and RTK GNSS survey is in process.
4
Address Offices in Brussels : Rue du Nord 76, BE – 1000 Bruxelles. Tel +32/2/217.39.72 Fax +32/2/219.31.47 E-mail: maurice.barbieri@clge.eu - www.clge.eu EU-Transparency Register of interest representatives - 510083513941-24
STUDENT CONTEST 2018 This object was surveyed only with DJI Phantom 4 Pro. The flight was planned with DroneDeploy software, which calculated the optimal trajectory. DJI flight took under 5 minutes. 139 photos were taken, of which we used 55. The height of the flight was 28 metres and ground pixel size was 6 millimetres. Data processing The coordinates that were collected with a GNSS device were imported to Civil 3D 2019 software. The contours and important breaklines of the stockpiles were drawn and then different surfaces were created for the bottom and the surface of the stockpiles. After that the heights of the two surfaces were compared to find the volume of the stockpile. The data collected by laser scanning was processed with Trimble Business Center and then converted into rcp format with Autodesk Recap. After that the point clouds were imported to Autodesk Civil 3D. GNSS base contour height was corrected using a control point and then the base surface of the stockpile was created. After that the surface of the stockpile was created from the points within the base contour. The point cloud was made more sparse to make the process easier for the computer. Distance between points was set to 5 centimetres and kriging method was used for creating the net. After that the heights of the two surfaces were compared to find the volume of the stockpile. Collected images were imported to Agisoft PhotoScan Professional. Images were oriented and then the coordinates of the GCP’s were also imported. The points were fixed manually on the images, which was fairly time consuming, especially when using two different sets of GCP’s. After fixing the GCP’s the images were oriented again. Then a point cloud was created from the images. For Aibot X6 images the data from the on-board IMU device was also used for orientation. The further processing of the point clouds was similar to processing the point clouds from Trimble SX10. 8 different models were created for Laiküla object and 4 different models for Karude object (Table 1). Table 1. Different models created by photogrammetric method Object
UAV
Used GCP’s
Laiküla peat extraction area
Aibotix Aibot X6
No GCP’s Aerosol paint GCP’s Photogrammetric GCP’s All GCP’s
DJI Phantom 4 Pro
No GCP’s Aerosol paint GCP’s Photogrammetric GCP’s All GCP’s
Karude gravel pit
DJI Phantom 4 Pro
No GCP’s 7 Photogrammetric GCP’s 9 Photogrammetric GCP’s 16 Photogrammetric GCP’s
Accuracy assessment of volumes For accuracy assessment, volume of different models was compared to the volume of the models based on laser scanning data. The model based on laser scanning was considered accurate for this study because it is considered more accurate then the other surveying methods used in this study. Both absolute and relative error was found for each model and it was observed if the relative error stays within the allowed error in Estonia. Similar to Richard Kramer Rhodes research (Rhodes 2017), root mean square errors (RMSE) were found for all the photogrammetric models. Gaussian RMSE formula was used for assessment of the accuracy of volume (Formula 1):
2 m n
(1) , (Randjärv 1997)
where Δ2 is the difference between arithmetic means of the volumes based on scanning data and volumes based on other methods and n is the number of different models.
Address Offices in Brussels : Rue du Nord 76, BE – 1000 Bruxelles. Tel +32/2/217.39.72 Fax +32/2/219.31.47 E-mail: maurice.barbieri@clge.eu - www.clge.eu EU-Transparency Register of interest representatives - 510083513941-24
5
STUDENT CONTEST 2018 To assess the accuracy of RMSE itself, RMSE of the result was found with the following formula (formula 2):
mm
m
(2), (Randjärv 1997)
2n 1
where m is the RMSE and n is the number of different models. In addition, time spent on different stages of the project and different methods was analysed. Results and conclusions The study found that the RMSE (Root Mean Square Error) of different photogrammetrical models of the Laiküla object (722,52 m3) was 14,31 m3 and the RMSE of different photogrammetrical models of the Karude object (674,04 m3) was 20,95 m3. Relative errors of all of the photogrammetrical models were under 4%, which is smaller than the allowed error of 12% in Estonia. Relative error of RTK GNSS model of Laiküla object was 5,81% and relative errors of different photogrammetric models of the same object were between 0,70% and 3,19% (Fig. 5). Respective errors of Karude object were 3,28% and 2,32% to 3,70% (Fig. 6). It was found that when measuring smaller objects it is efficient to use RTK GNSS, but the advantages of UAV photogrammetry become apparent as the size and complexity of the objects grows.
RELATIVE ERROR(%)
Relative error of volume compared to laser scanning based volume 4.00 3.00 2.00 1.00 0.00 -1.00 -2.00 -3.00 -4.00 -5.00 -6.00 -7.00 Ilma tähisteta Trimble R4-3 GNSS vastuvõtja Suhteline erinevus (%)
-5.81
Fotogram Aerosoolv mKõik ärviga meetrilise tähised tähised d tähised
Ilma tähisteta
Aibotix X6 -2.44
-1.04
1.07
Fotogram Aerosoolv mKõik ärviga meetrilise tähised tähised d tähised DJI Phantom 4 Pro
-0.70
-1.98
-3.19
2.73
1.10
Fig. 5. Relative error of volume of different models of Laiküla object
Results show that using UAV photogrammetry to determine stockpile volumes is sufficiently accurate with both of the tested UAVs. Still, it is important that using UAVs for geodetic work, would be done by trained professionals and all the legal requirements would be met. It was found that sufficient accuracy was also achieved without using the GCP’s (Ground Control Points). However, using GCPs did increase the accuracy. Furthermore, using different types of GCPs had no impact to the accuracy of the volume. Still, using more GCPs also increased the accuracy. It was found that results of the two objects with different characteristics were in the same accuracy range, however results of the regularly shaped peat stockpile were a bit more accurate. Results show that, when surveying large areas, using UAV photogrammetry can save a lot of time on field work compared to GNSS surveys.
6
Address Offices in Brussels : Rue du Nord 76, BE – 1000 Bruxelles. Tel +32/2/217.39.72 Fax +32/2/219.31.47 E-mail: maurice.barbieri@clge.eu - www.clge.eu EU-Transparency Register of interest representatives - 510083513941-24
STUDENT CONTEST 2018
RELATVIE ERROR (%)
Relative error of volume compared to laser scanning based volume 5.00 4.00 3.00 2.00 1.00 0.00 -1.00 -2.00 -3.00 -4.00 Ilma tähisteta Trimble R4-3 GNSS vastuvõtja Suhteline erinevus (%)
-3.28
7 fotogrammmeetrilist tähist
9 fotogrammmeetrilist tähist
16 fotogrammmeetrilist tähist
DJI Phantom 4 Pro 3.70
-3.23
-3.03
-2.32
Fig. 6. Relative error of volume of different models of Karude object
On this basis, it can be concluded that UAV photogrammetry could be on efficient alternative for determing stockpile volumes in Estonia. It can also be concluded that even the commercial UAVs could be used for tasks that don’t require very high accuracy.
References Kokamägi, K. 2018. Mehitamata õhusõiduki abil tehtud aerofotode põhjal puistangu mahtude arvutamise täpsus: Master’s thesis. Estonian University of Life Sciences, Institute of Forestry and Rural Engineering. Tartu. Markšeideritöö kord. 2012. Riigi Teataja. Estonian legislation. [cited 15 March 2018]. Available from Internet: https://www.riigiteataja.ee/akt/125012012004 Randjärv, J. 1997. Geodeesia I. Tartu. Rhodes, R., K. 2017. UAS as an Inventory Tool: A Photogrammetric Approach to Volume Estimation: Master’s thesis. University of Arkansas. Monticello. [cited 3 April 2018]. Available from Internet: http://scholarworks.uark.edu/cgi/viewcontent.cgi?article=3963&context=etd Web mapping service of Estonian Land Board [online]. 2018 [cited 2 May 2018]. Available from Internet: http://xgis.maaamet.ee/xGIS/XGis
Address Offices in Brussels : Rue du Nord 76, BE – 1000 Bruxelles. Tel +32/2/217.39.72 Fax +32/2/219.31.47 E-mail: maurice.barbieri@clge.eu - www.clge.eu EU-Transparency Register of interest representatives - 510083513941-24
7