You are on page 1of 8

UAV photogrammetry to analyse the accuracy of Digital Elevation Model (DEM) in a

Case Study

Author 1---
Author 2---
Author 3---

Abstract
Geodetic survey of the terrain surface is generally carried out by use of conventional surveying methods and
instruments. With the new technology like UAVs and their combination with digital cameras, new fanged
approaches are experimented in documentation of earth’s surface, DEM is one of them. It is a quantitative
representation of earth’s terrain. The DEM can be generated using interferometry, ground and laser surveying (like
LiDar), photogrammetry etc. Each DEM carry intrinsic errors due to data acquisition technology (UAV, GPS, Digital
Camera) and processing methodology in relation to particular terrain (here mines). The accuracy of these data sets
are often unknown and non-uniform.

The primary scope of the paper is to gauge the performance of an UAV consolidated with camera for geodetic
application by the way of DEM of an open-pit mine. For this objective 20 checkpoints i.e. GCPs were surveyed and
approximatively 650 raw images were captured with GoPro HERO 6 camera. Over 20 million cloud points were
used in DEM generation process. Accuracy was evaluated with 20 checkpoints and overall accuracy obtained was
0.02642 m or 2.642 cm (RMS error). The slope curvature and terrain morphology strongly influences the DEM
accuracy which is also discussed here. The outcomes and results of the study shows that Data procured from UAV
photogrammetry have befitting accuracy and advantages like minimal cost, nominal time consumption, minimal
manpower or fieldwork etc. So, it is possible to extensively use this technique for surveying, map producing or other
engineering or mining application.

1. Introduction Vehicles (UAVs) are used to pass this handicap. Nowadays, the use of
UAVs is increasing day by day due to its advantages at cost, inspection,
Digital Elevation Model (DEM) is an important topographic surveillance, recon-naissance, and mapping [2].
product and essential demand for many applications. Traditional
methods for creating DEM are very costly and time consuming because The term UAV is used commonly in the computer science and artificial
of land surveying. In time, Photogrammetry has become one of the intelligence community, but terms like Remotely Piloted Vehicle (RPV),
major methods to generate DEM. Recently, airborne Light Detection Remotely Operated Aircraft (ROA), Remote Controlled (RC) Helicopter,
and Ranging (LiDAR) system has become a powerful way to produce Unmanned Vehicle Systems (UVS), and Model Helicopter are often in
a DEM due to advantage of collecting three-dimensional information use [3]. The RC and Model Helicopters are clearly defined by the
very effectively over a large area by means of precision and time [1]. Unmanned Vehicle Systems International Associations as mini, close,
However, the main disadvantage of aerial manned platforms such as short, and med-ium range UAVs depending on their size, endurance,
air-planes is being expensive, especially for small study areas. During range, and flying altitude [4,2]. UAVs are to be understood as uninhabited
the last decades, low-cost Unmanned Aerial and reusable motorized aerial vehicles which are remotely controlled,
semi-autonomous or have a combination of these capabilities, and that
can carry various types of payloads, making them capable of performing
specific tasks within the earth’s atmosphere, or beyond,
for a duration, which is related to their missions [5]. The UAVs system was essentially depends on both batty and payload weight. A lot of
firstly applied for military goals but in 1979 Przybilla and Wester-Ebbinghaus features of this model are available such as Altitude Hold, GPS Hold,
[6], applied one of the earliest experiments with fixed wing UAVs for geo- CareFree, Coming Home, Fail Safe, Low Battery Protection, Auto
matic applications. Take Off and Landing, Waypoint Flight and Follow Me. Mikrocopter
(MK Tools) software let us to view the navigation and flight status
UAVs are established as serious alternative for traditional data capture, information in real time. It is possible to perform autonomous flight
especially while aiming at mapping application with high spatial and temporal plan over the online maps. Moreover, some details such as horizontal
resolution and introduces also a low-cost alternative to the classical manned and vertical speed, altitude, direction, waiting time at willing points,
aerial Photogrammetry [7,8]. Data collection just requires small, light UAV coordinate information, and camera angle are also can be specified.
platform with a control system and standard consumer grade digital camera Waypoint Flight electronic is capable of autonomous flight in a 250
[9]. m radius area and 250 m fly height for a standard 32-points route.

Basically UAVs applications can be classified as forestry and agriculture, The digital camera was GoPro Hero 6 It has featured with 18
archaeology and cultural heritage, environmental surveying, traffic Megapixel and 22.3 ⁄ 14.9 CMOS image sensor. Single, continuous,
monitoring, 3D reconstruction [2]. There are various UAVs based civilian and self-timer drive abilities are among the digital camera features.
applications at the scientific literature [8,10–16]. In this study UAV plat-form The Body weight of the device is 298 g and it reaches a total the
is used to capture aerial images of Sahitler hills in purpose of generating the weight of 403 g with EF-M 22 mm lens.
digital elevation model.

2. Study area and equipment 3. Application

. 3.1. Flight plan and data collection

2.1. UAV, digital camera and ground control boards It was difficult to prepare a flight plan with the available software
for a UAV surveying applications. However, MK Tools allow us to
The OCTO XL UAV was used to capture images. It uses the microcopter pass this obstacle. The main advantage of this software is to require
electronic. The weight of it is approximately 2.05 kg and the payload is 2 kg. minimum user interferences. After some parameters such as
Flying time overlapping of photos,
waypoints, flight path, and altitude had been defined on an online map eliminating distorted or blurred ones. We used 200 selected photos.
and uploaded, the UAV carried out the flight plan automatically. It is The interior orientation of photos was determined in Photo Modeller
necessary to remember some pre-cautions: weather condition of the software. The process proceeds with aligning photos, building
study area, daylight sta-tus for better photos, backup battery depends geometry, and texture building for realistic appearance. The software
on the study area, and controls of flight, engine, and navigation of generates an assessment report for each step of processing.
UAV against any errors.
For a geomatic application, it is necessary to georeference the
When all the necessary controls had finished, online maps were data. There are two common methods for that. Direct georeferencing
uploaded to OSD. The flight plan was prepared based on this map depends on the simultaneity of GPS time and the camera inertial time.
with a 4 column and 32 point route and the altitude were 60 m (Fig. After the adjustment of these data, the 3D georeferenced point cloud
1) and was uploaded to the UAV. Then automatic flight was started. can be generated directly. At the indirect technique, the Ground
Control Points (GCP) that were measured before the flight by the
At each waypoint the UAV captures photo, but the cam-era was method of RTK GPS surveying are used to georeferenced the data.
designed for capturing photos in a 2 s interval in purpose of many The color and size of GCPs must be suitable to distinguish at the
more photos. This has given us the possibility of eliminating damaged natural color of study area due to marking them during the process. In
or blurry photographs. Moreover, it is still possible to cut in during this application, indirect method was used to georeferenced with 27
the flight for any reason. When the flight over, the UAV automatically GCP which were measured by RTK with a base station top of the
turns where it takes off. At the end of the flight the UAV collected Afyon Kocatepe University Faculty of Engineering faculty about 3
250 photos of the study area. km away from the study area at ITRF96 datum and 2005.00 Epok.
The accuracy of GCPs is calculated as 2.37 cm.

3.2. Data processing and DEM generation


Georeferenced 3D point cloud was generated as both sparse point
The main aim of process is to produce a georeferenced 3D point cloud as 4 092 485 points and dense point cloud as 26 025 883 points
cloud by handling with irregular and overlapping aerial image data (point density 450 pt per m2). Point cloud and coloured point cloud
[17]. Existing software’s can generate a 3D point cloud such as; are shown in Fig. 2.
Agisoft PhotoScan (commercial software) that has been used in this Finally, DEM was generated from dense point cloud at a
study (URL-1). resolution of 5.203 cm per pixel (Fig. 3).
The software is advanced in UAV applications and allows to
generate DEM and orthophoto in a willed coordinate system. For a
full performance of software, it’s recommended to use a powerful
computer due to the huge amount of data [17].

3.3. Accuracy analysis

Accuracy analysis was performed with the sample point surveying


in the study area and comparing them with generated DEM. In this
purpose 30 check points was measured
Fig. 3. Generated DEM of the study area.

Fig. 4. Measured errors between generated DEM and checkpoints


Table 2
Error in Cartosat DEM with respect to GCPs.

GPS point UTM coordinates Z Z Error in DEM


(m)
GCPs (m) DEM (m)

X Y
Point 1 769,986.27 3,367,905.01 454.65 456.28 1.63
Point 2 777,381.85 3,369,567.47 648.50 643.32 −5.18
Point 3 771,532.48 3,370,524.05 492.87 494.12 1.25
Point 4 777,340.91 3,358,636.80 490.26 484.76 −5.5
Point 5 770,981.69 3,360,624.53 480.39 482.69 2.3
Point 6 772,976.63 3,360,482.88 467.71 464.66 −3.05
Point 7 768,903.00 3,345,906.56 371.58 373.54 1.96
Point 8 765,861.25 3,347,501.29 371.40 374.57 3.17
Point 9 783,589.54 3,368,403.41 775.17 768.77 −6.4
Point 10 772,341.35 3,343,955.79 364.73 366.36 1.63
Point 11 770,017.87 3,364,620.82 433.73 437.28 3.55

both the elevation was shown in Table 3. The RMSE of DEM calculated is 6.3. ASTER and SRTM height comparison with SOI Toposheet
4.83 m and the mean error is 0.19 m.
Accuracy of ASTER and SRTM DEMs was also evaluated using 30 point
6.2. Comparison of ASTER and SRTM with GCPs location taken from SOI Toposheet height (BM/Spot height). The vertical
datum of SOI Toposheet is MSL which is closely com- parable with EGM96.
The ASTER GDEM and SRTM data were compared with 11 DGPS derived
GCPs. The GPS system provides elevation reference to WGS84 surface but Therefore, matching datum is not necessary. The absolute differences in
ASTER and SRTM are referenced to EGM96 sur- face. These two heights heights are shown in Fig. 6. The RMS error calculated is 16.06 m and 18.91
cannot be compared directly. The GPS based ellipsoid height was converted −
m with mean bias of 2.14 m and 2.10 m for ASTER and SRTM, respectively.
into EGM96 geoid reference sur- face and compared with ASTER and SRTM ASTER is showing negative bias which suggests the under estimation of
height shown in Table 4. The RMS error for the ASTER and SRTM calculated elevation values but SRTM gives the positive bias which leads to over estima-
is 6.08 m and
tion. The RMS error is much higher compared to error specified in the
9.2 m with mean error of 2.58 m−and 2.94 m, respectively.
− In both cases, the
validation report of the nodal agency. The level of agreements (R2 ) between
error is less than the error specification given by the nodal agency (8.86 m for
ASTER and 16 m for SRTM). This could be because the GCPs are taken in the Toposheet height and ASTER and SRTM are 0.982 and 0.976 (Fig. 6c).
flat terrain due to accessibility of the field work. Correlation between ASTER and Toposheet height is higher which signifies
that accuracy of ASTER is better compared to SRTM.

Table 3
Error in DEM with respect to derived height.

BM/spot height GCP height DEM Height Geoid undulation Transform height Error in height

1 489.1 447.99 −44 491.99 2.89


2 544 495.5 −44.44 539.94 −4.06
3 618 575.47 −44.15 619.62 1.62
4 601.2 551.84 −44.63 596.47 −4.73
5 638.8 597.22 −45.02 642.24 3.44
6 453.8 405.68 −46.84 452.52 −1.28
7 468 419.54 −44.57 464.11 −3.89
8 482 441.24 −44.32 485.56 3.56
9 561.4 522.22 −44.29 566.51 5.11
10 675 631.77 −45.55 677.32 2.32
11 713 664.37 −46.06 710.43 −2.57
12 774 723.02 −46.55 769.57 −4.43
13 515 468.58 −46.68 515.26 0.26
14 418.6 375.23 −46.15 421.38 2.78
15 410 359.94 −45.21 405.15 −4.85
16 543 493.77 −45.09 538.86 −4.14
17 581 534.79 −45.22 580.01 −0.99
18 579 538.18 −45.18 583.36 4.36
19 848 789.89 −46.6 836.49 −11.51
20 672 640.6 −46.12 686.72 14.72
21 818 774.08 −45.98 820.06 2.06
22 468 419.54 −44.56 464.1 −3.9
23 478 439.47 −44.14 483.61 5.61
24 695 653.34 −43.27 696.61 1.61
25 631 582.97 −44.04 627.01 −3.99
26 705 658.05 −45.77 703.82 −1.18
27 415 373.37 −47.03 420.4 5.4
28 410 358.94 −47.21 406.15 −3.85
29 624 579.75 −45.31 625.06 1.06
30 491 450.55 −44.72 495.27 4.27
Table 4
Error in ASTER and SRTM DEM with respect to GPS derived height.

GPS point Z Geoid undulation Transform GPS height Z Error Z DEM Error
GCPs (m) DEM (m) DEM (m) (m) DEM (m)

Point 1 454.65 −42.89 497.54 488 −9.54 490 −7.54


Point 2 648.50 −42.10 690.60 695 4.40 692 1.40
Point 3 492.87 −42.40 535.27 526 −10.27 531 −4.27
Point 4 490.26 −43.86 534.12 521 −9.12 530 −4.12
Point 5 480.39 −43.93 524.32 514 −2.32 521 −3.32
Point 6 467.71 −43.81 511.52 504 −8.52 503 −8.52
Point 7 371.58 −46.08 417.66 420 −17.66 412 −5.66
Point 8 371.40 −46.03 417.43 422 4.57 410 −7.43
Point 9 775.17 −41.74 816.91 804 −12.91 828 11.09
Point 10 364.73 −46.18 410.91 424 −10.91 414 3.09
Point 11 433.73 −43.39 477.12 483 3.88 474 −3.12

less than 10 m elevation error. 30% areas have greater than 15 m elevation error
6.4. ASTER and SRTM height comparison with Cartosat DEM
where variation of relief is higher.
The elevation difference surface of SRTM ranges from 142 m to − 206 m,
ASTER and SRTM height value was also compared with Car- tosat
though maximum pixels fall under 25 m to 12 m. The RMS error for the
height (reference vertical datum converted into EGM96) on the same 30- −
surface is 17.76 m with mean bias of 5.42 m. It was found from the difference
point location mentioned in the previous section. The absolute differences −
surface that positive bias is associ- ated with vegetation covered
in heights are shown in Fig. 7. The RMS error calculated for ASTER and mountainous area and negative bias associated with open land and other
SRTM are10.68 m and 13.18 m with mean bias of 3.09 and 2.37 m, land cover types. Fig. 8 shows in 69% of the study area; the elevation error is
respectively. The level of agreements (R2) between Cartosat height with less than 15 m which satisfy the error specification of the SRTM mission. 31%
ASTER and SRTM are 0.992 and 0.988. The assessment of a DEM surface area with error greater than 15 m is mainly due to rough terrain. The his-
togram of the error surface of ASTER and SRTM also shows a normal
with respect to few sample points is not good enough. Due to this reason,
distribution (Fig. 10) which is statistically significant.
surface-to-surface comparison of ASTER and SRTM with Cartosat DEM was
carried out. Cartosat does not provide a perfect elevation model. But 6.5. Effect of terrain morphology in DEM accuracy
Cartosat DEM is considerably more accurate compared to ASTER and SRTM.
Hence, Cartosat DEM is considered as reference data. To match the vertical Terrain morphology is one of the major influencing factors for vertical
datum, Cartosat surface was transferred to EGM96 refer- ence datum. The accuracy of DEM. In order to evaluate this, Cartosat DEM is divided into 5
altitudinal zones (<400 m, 401–500 m, 501–600 m,
height differences between WGS84 and EGM96 for the entire study area
(Fig. 8) have a mean value of 44.71 with minimum and maximum of 46.71
− converted
and 41.42. Using this offset, Cartosat DEM ellipsoid height was
into EGM96 geoid ref- erence −surface (refer
−Eq. (1)) and the minimum,
maximum, mean and standard deviation of the surface were changed to
321, 1015, 596 and 110. ASTER and SRTM were then compared withCartosat
and different surfaces were generated.The ASTER elevation difference
surface ranges from 138 m to 199 m, but major area comes under −30 m to
15 m. The overall RMS error for the surface is 12.62 with mean bias of −5.53
m. Mean bias
The negative bias is associated with mountain ridge, plain land and vegetated
area. The aerial distribution height error shown in Fig. 9

You might also like