Page 93 - NGTU_paper_withoutVideo

Modern Geomatics Technologies and Applications

EVALUATING THE ACCURACY OF CLUSTERING TECHNIQUES FOR
LOCATING GROUND CONTROL POINTS IN UAV PHOTOGRAMMETRY
PROJECTS: SOME PRELIMINARY RESUALTS

1*
1
2
Fatemeh Bakhshi , Masood Varshosaz , Saied Pirasteh , Kamyar Hassanpour 1
1 Geomatics Engineering Faculty, K.N. Toosi University of Technology, Valiasr St., Tehran, Iran
2 Department of Surveying and Geoinformatics, Faculty of Geosciences and Environmental Engineering,
Southwest Jiaotong University (SWJTU), Chengdu, China
* varshosazm@kntu.ac.ir

KEY WORDS: UAV, Clustering, Ground Control Points, Evaluation

ABSTRACT: Typically, in order to achieve the accuracy of a UAV photogrammetric project, it is necessary to measure a
large number of ground control points. Capturing this number of points is a time-consuming and costly process, especially
in mountainous and hilly areas. To reduce the number of control points, there exist several techniques that suggest optimal
locations of the ground control points. Yet, due to the complexity of ground, most of these techniques can lead to either
redundant or missing locations required for a project to be oriented properly. Therefore, in practice, usually, the location
of the points is determined either by visiting the ground in person or by visiting the area of interest on Google Earth.
Aiming to automate the process, we decided to is to examine the ability of clustering methods in selecting an optimal, and
perhaps minimum, set of ground control points. In this study, we used 3D distance as the clustering parameter. Taking
determined cluster centres as optimal locations can provide even distribution for the location ground control points. For
this, the Partitional, Hierarchical, Model, Fuzzy, Graph, and Modern clustering techniques were evaluated. The results
showed that the Average Linkage, SOM and FCM are the best, from the consistency and accuracy points of views. It was
also noted that the points selected using these three techniques, in addition to being less, provide models more accurate
than those obtained using ground control points that are selected manually by the operator.

1. Introduction
Ground control points (GCPs) are one of the factors that make a significant contribution to the accuracy of an Unmanned
Aerial Vehicle (UAV) photogrammetry project [1]. Ablanedo [2] showed that the number of control points has a direct effect on
the accuracy of a photogrammetric project. The accuracy ranges from 2 to 5 times the ground sampling distance (GSD). The
number of these points is often high, due to the use of non-metric cameras which usually suffer from poor geometry. As a result,
the cost and time of capturing points will be significant, especially if the area is hilly or mountainous. Therefore, it is necessary
to develop a method that uses a smaller number of points to ensure the accuracy of the required project.
So far, numerous efforts have been made to optimise the number and location of ground control points. In [3], error
minimisation was introduced as a criterion for determining the number of optimal ground control points. In [4], the effect of the
distribution of ground control points on ortho-mosaic accuracy was investigated. In this study, the distance between the ground
control points was equal to 10% of the area width. Harvin and Lossier [5] stated that the number and distribution of ground
control points play a major role in the accuracy of the district model. Based on their evaluations, the optimum distance between
the points of control is 1.5 times the flying height. Also, ground control points should be closer to the steep ground. In [6], to
determine the size and motion of the glaciers, the effect of the number and distribution of ground control points in the digital
surface model has been analysed. In this study, it has been shown that, on average, local precision is reduced by a distance of
100 meters from the nearest ground control point. In [1], to determine the accuracy of photogrammetric products, six GCP
distribution designs of were selected (Figure 1) and the accuracy of the region model was evaluated. They state that in more than
97% of the times, the effective parameter in photogrammetric results relates to the number and distribution of ground control
points. To reduce the number of control points, we need to identify optimal locations of the ground control points. After a
thorough study, Martínez-Carricondo [7] suggest placing GCPs around the edge of the study area to minimise the planimetric
errors. They also, propose the creation of a stratified distribution of GCPs inside the study area with a density of around 0.5–1
GCP × ha−1 to minimize the height error.
Unfortunately, due to the complexity of the ground, most of these techniques can lead to either redundant or missing
locations required for a project to be oriented properly. Therefore, in practice, usually, the location of the points is determined
either by visiting the ground in person or by visiting the area of interest on Google Earth. In a similar effort, Li [8], studied the
ability of clustering techniques in choosing the number and distribution of ground control points for georeferencing of remote

1