See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/369362467

Drone Image Processing Using Pix4D - Orthophoto, DEM and Overlay

Technical Report · March 2023

DOI: 10.13140/RG.2.2.30527.38567

CITATION READS
1 2,123

1 author:

Bipin Paudel
Tribhuvan University
6 PUBLICATIONS 1 CITATION

SEE PROFILE

All content following this page was uploaded by Bipin Paudel on 19 March 2023.

The user has requested enhancement of the downloaded file.

 Drone Image Processing Using Pix4D
Bipin Paudel bipinschwimmer389@gmail.com

ABSTRACT

The report on "Drone Image Processing Using Pix4D Software" discusses the
use of drone imagery for data collection and analysis. Pix4D software is
popular tools for processing drone. The report includes the technical aspects
of using these software packages and highlights their benefits and limitations..
The report concludes by emphasizing the importance of proper data
processing and interpretation to ensure accurate and reliable results. Overall,
the report provides a comprehensive overview of the use of drone imagery
and the software tools used for processing it. The drone images are taken over
Paschimanchal Campus Pokhara using Drone camera with Camera Model
Name(s) FC300S_3.6_4000x3000 (RGB) and the image covering 0.08 sq. mi.
/ 54.3791 acres of area with Average Ground Sampling Distance (GSD) 1.88
cm / 0.74 inch.

Keyword:
Orthomosaic, DTM, DSM, GSD

vi
 Contents
ABSTRACT .................................................................................................. vi
List of Figures .............................................................................................. viii
List of Tables ................................................................................................. ix
List of Abbreviations .......................................................................................x
1. INTRODUCTION .....................................................................................1
2. SOTWARE USED FOR DRONE IMAGE ANALYSIS .........................1
3. METHODOLOGY AND STUDY AREA ................................................2
3.1 STUDY AREA ......................................................................................2
3.2 METHODOLODY ................................................................................4
3.2.1 Initial Processing: ............................................................................4
3.2.2 Point Cloud and Mesh: ....................................................................5
3.2.3 DSM, Orthomosaic and Index: .......................................................6
4. RESULT ....................................................................................................7
4.1 ORTHOPHOTO ....................................................................................7
4.2 DIGITAL SURFACE MODEL.............................................................9
5. ANALYSIS AND DISCUSSION ...........................................................10
6. CONCLUSION .......................................................................................12
7. RECOMMENDATION ..........................................................................12

vii
 List of Figures
Fig 1 : Pix4D .................................................................................... 2
Fig 2 : Study Area ............................................................................ 3
Fig 3 : Flight Plan, Top view of the initial image position .............. 4
Fig 4 : Computed Image and showing uncalibrated image position8
Fig 5: Orthophoto ............................................................................. 8
Fig 6 : Orthophoto Overlay with OSM ............................................ 9
Fig 7 : Digital Surface Model......................................................... 10
Fig 8 : Number of overlapping images computed for each pixel of
the orthomosaic .............................................................................. 11

viii
 List of Tables
Table 1 : Initial processing Details ..................................................................5
Table 2 : Point Cloud Densification Details ....................................................6
Table 3 : DSM, Orthomosaic and Index Details .............................................7
Table 4 : Absolute Geolocation Variance .....................................................11

ix
 List of Abbreviations

DTM Digital Terrain Model

DSM Digital Surface Model

GCP Ground Control Point

x
 1. INTRODUCTION

DEM stands for Digital Elevation Model, which is a representation of the

topographic features of the earth's surface. In other words, a DEM is a map of
the shape and height of the land, similar to a relief map. A DEM is created
using information gathered from various sources, such as satellite imagery,
aerial photography, and ground-based surveys. This data is processed and
combined to create a three-dimensional model of the earth's surface. A DEM
is used for a variety of purposes, such as in mapping and planning, in
environmental studies to track changes in the landscape over time, in geology
to study the shape of the land and how it has changed, and in civil engineering
to help design and build infrastructure projects such as roads and bridges. In
simple terms, a DEM is like a digital map that shows the shape of the land and
its elevation. It helps us understand the land and how it can be used, and it
helps us make informed decisions about the environment and infrastructure.
Drones are used in surveying to create detailed maps of land and identify
changes to terrain, water levels, and vegetation. These maps are used in
various fields such as construction, urban planning, and agriculture. To take
images with a drone, a surveyor first identifies the area they want to map and
plans out a flight path for the drone. The drone is equipped with a high-
resolution camera that captures images as it flies over the area. These images
are then stitched together using software to create a 3D model of the area.
Drones have many advantages over traditional surveying methods. They can
cover large areas quickly and with high accuracy, reducing the need for
ground-based surveys. They are also less expensive and more
environmentally friendly than traditional methods, as they require less
equipment and do not disturb the land. Overall, the use of drones in surveying
is a valuable tool that provides detailed, accurate data for various industries.

2. SOTWARE USED FOR DRONE IMAGE

ANALYSIS
There are several software options available for drone image analysis. Here
are a few commonly used ones like Agisoft, Pix4D, Global Mapper, etc.

1
 Fig 1 : Pix4D

Pix4D: Pix4D is a photogrammetry software that facilates us to create 3D

models, maps, and survey data from aerial and ground images. It works by
taking a series of overlapping images of an area or object from different angles
and using computer algorithms to process these images into a detailed 3D
model or map. With Pix4D, users can process their images to generate
orthomosaics, point clouds, 3D models, and more, using advanced
photogrammetry techniques. The software is commonly used in a wide range
of industries, including agriculture, construction, mining, and surveying, to
provide accurate and up-to-date information about land and structures.

3. METHODOLOGY AND STUDY AREA

3.1 STUDY AREA
Paschimanchal Campus is located in the city of Pokhara, which is situated in
the western part of Nepal. The campus itself is located in the Lekhnath
Municipality of Pokhara, on the western side of the city. It is easily accessible
from the main highway and spread area over 0.159 square kilometers.

2
 Fig 2 : Study Area

Initial position of the images, ie the flight plan is shown below. As shown in
figure different flight plan are taken. he green line follows the position of the
images in time starting from the large blue dot.

3
 Fig 3 : Flight Plan, Top view of the initial image position

3.2 METHODOLODY
PIX4Dmapper divides the image processing into three major steps. They are:
3.2.1 Initial Processing:
PIX4Dmapper first computes key points on the images. It uses these
key points to find matches between the images and prepares the quality
report of this step. The initial processing step involves the following
sub-steps:
a) Image Import: The first step is to import the images into the software.
This can be done by selecting the folder containing the images.
b) Calibration: The software then analyzes the images to determine the
internal and external camera parameters, such as focal length, principal
point, and lens distortion.
c) Image Orientation: The software uses the camera parameters to
determine the position and orientation of each image.
d) Image Matching: The software then matches common features
between images to create tie points, which are used to align the images.

4
 e) Point Cloud Generation: Finally, the software generates a 3D point
cloud from the aligned images.
It almost took 43 minutes for initial processing of images available.

Table 1 : Initial processing Details

Above figure gives information on hardware system information and

its operating system. Also the coordinate system is WGS 1984/ UTM
Zone 44N.

3.2.2 Point Cloud and Mesh:

This step increases the density of 3D points of the 3D model computed
in step 1 which leads to higher accuracy both for the DSM and the
Orthomosaic. The point cloud and mesh step involves the following
sub-steps:
a) Point Cloud Cleaning: The software removes any spurious or outlier
points from the point cloud.

5
 b) Point Cloud Classification: The software classifies the points based
on their location and characteristics, such as ground, vegetation,
buildings, and so on.
c) Digital Surface Model (DSM) Generation: The software generates a
DSM from the classified points.
d) Mesh Generation: The software generates a mesh surface from the
point cloud and DSM.

Table 2 : Point Cloud Densification Details

3.2.3 DSM, Orthomosaic and Index:

This is the final step for creating the desired outputs for DSM and
orthomosaic.
The DSM, orthomosaic and index step involves the following sub-
steps:
a) DSM Editing: The software allows users to edit the DSM by
removing objects, filling holes, or smoothing the surface.

6
 b) Orthomosaic Generation: The software generates an orthomosaic by
projecting the images onto the DSM and blending them together.
c) Index Calculation: The software calculates various indices, such as
NDVI (Normalized Difference Vegetation Index), NDRE (Normalized
Difference Red Edge Index), and so on, using the orthomosaic.
Finally, the software allows users to export the results in various
formats, such as GeoTIFF, LAS, OBJ, and so on, for further analysis.

Table 3 : DSM, Orthomosaic and Index Details

4. RESULT
4.1 ORTHOPHOTO
The orthophoto below is the processed output of 1089 drone images which
took around 6 hours for orthomosaic generation out of which 1080 images
(99%) are calibrated. There is 1.61%relative difference between initial and
optimized internal camera parameters. Out of the orthomosaic developed we
clipped out the unnecessary area. The median of 1957.15 matches per
calibrated image. The number of calibrated images is 1080 out of 1089 and
geolocated images is 1089 out of 1089.
The image below represents 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). Red dots indicate disabled or uncalibrated images. Dark
green ellipses indicate the absolute position uncertainty of the bundle block
adjustment result.
7
 Fig 4 : Computed Image and showing uncalibrated image position

Fig 5: Orthophoto

For Georeferencing we don’t use 3D GCPs but used 2D GCPs which we

manually take from google map for four points. In QGIS we use Open Street
Map as base layer and perform map overlay with the georeferenced
orthophoto we obtained. And orthophoto comes to almost fit at the position

8
of Paschimanchal campus in OSM . However slight deviation result that
depend on with how much accuracy and how many do we take the reference
points for georeferencing orthophoto.

Fig 6 : Orthophoto Overlay with OSM

4.2 DIGITAL SURFACE MODEL

The DSM is created with resolution of 1 xGSD(1.88 [cm/pixel]). In DSM
filter, noise filtering and surface smoothing is applied with type sharp. Raster
DSM is created with method of Inverse Distance Weighting while merging
tiles. It took almost 1. 5 hours for creation of DSM while processing through
pix4D.

9
 Fig 7 : Digital Surface Model

5. ANALYSIS AND DISCUSSION

1089 images taken from DJI Phantom drone was processed for almost 9 hours
using Pix4D. Single position in the orth ophoto is formed by overlapping of
more than 5 images. The figure below shows 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.

10
 Fig 8 : Number of overlapping images computed for each pixel of the orthomosaic

The Mean Reprojection Error [pixels] is found to be 0.141. The number of

2DKeypointObservations for Bundle Block Adjustment is 2222294. The
number of 3D points for Bundle Block Adjustment is 720433
.

Table 4 : Absolute Geolocation Variance

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
11
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. The image geolocation errors do not correspond
to the accuracy of the observed 3D points. Since 2D GCPs is taken for
georeferencing the geolocation error in Z axis seems to have more
uncertainity. In order to reduce this error, it is recommended to use a drone
with a stable gimbal and acquire UAV images in optimal weather conditions,
use GCPs take from DGPS survey rather than taking location data from
uncertain sources, also we can extend the logging time of DGPS for better
accuracy.

6. CONCLUSION
Using UAVs for data gathering enables efficient and accurate data collection,
resulting in the creation of high-resolution orthophotos of the Pashchimanchal
campus. The orthophotos offer a comprehensive and precise representation of
the campus and are useful for designing, managing assets, and monitoring
changes over time. The effectiveness and efficiency of UAV technology for
mapping and surveying have been demonstrated by the successful generation
of the orthophoto of the campus.

7. RECOMMENDATION
The orthophoto production project that utilized UAV technology was
completed successfully and demonstrated the potential of this technology for
mapping and surveying purposes. For future projects that require efficient and
precise data collection, the use of UAVs is recommended. But make sure of
condition of the devices and instruments used. To ensure the orthophoto's
accuracy and reliability, it is crucial to implement quality control measures,
such as image distortion checks and the use of multiple ground control points.
GCPs taken with high logging time period is recommended. A sufficient
number of GCPs is essential for achieving a highly accurate orthomosaic, and
having a better front and side overlap is recommended. Computer with better
processing power is recommended so that the processing would not terminate
at middle of process.

12
 13

View publication stats