doi:10.6062/jcis.2015.06.01.0101 Cruz, Shiguemori & Guimar˜ aes 121
J. Comp. Int. Sci. (2015) 6(3):121-136
http://epacis.net/jcis/PDF_JCIS/JCIS-0101.pdf
jcis@epacis.net
©2015 PACIS (http://epacis.net)
PAN-AMERICAN
ASSOCIACION OF
COMPUTATIONAL
INTERDISCIPLINARY
SCIENCES
A comparison of Haar-like, LBP and HOG approaches to concrete and asphalt
runway detection in high resolution imagery
Juliano E. C. Cruz
a1
, Elcio H. Shiguemori
b2
and Lamartine N. F. Guimar˜ aes
c3
a
CAP,National Institute for Space Research (INPE), S˜ ao Jos´ e dos Campos, SP, Brazil
b
EGI-S, Institute for Advanced Studies (IEAv), S˜ ao Jos´ e dos Campos, SP, Brazil
c
ENU, Institute for Advanced Studies (IEAv), S˜ ao Jos´ e dos Campos, SP, Brazil
Received on october 11, 2015 / accepted on december 20, 2015
Abstract
In this article, the three most used object detection approaches, Linear Binary Pattern cascade, Haar-
like cascade, and Histogram of Oriented Gradients with Support Vector Machine are applied to automatic
runway detection in high resolution satellite imagery and their results are compared. They have been
employed predominantly for human feature recognition and this paper tests theirs applicability to runways.
The results show that they can be indeed employed for this purpose with LBP and Haar approaches
performing better than HOG+SVM.
Keywords: runway detection, satellite imagery, boosting, boosted cascade, LBP, Haar-like, HOG, SVM, com-
putational mathematics.
1. Introduction
Experts can usually recognize most of the targets in satellite imagery with high spatial resolution. Some
objects are easy to be visually recognized; others however, require a high level of expertise. Due mainly to
automatic object detection and identification evolution, today it is possible, to aid human operators, or even
substitute them, in tasks dealing with images or videos [1, 2].
There is a wide range of applications that can employ some sort of automatic detection or identification
procedures. UAV (Unmanned Aerial Vehicle) platforms are one of them. They are currently widely in use
for military, police, and civilian applications [3].
Automatic runway recognition can be an important task for UAVs. They can use it for landing, for air
strikes, or even for self-localization procedures [4]. The best runways to perform such procedure are made
of concrete and asphalt. They can be found, generally, in medium to large cities and air force bases.
UAVs can have fully-autonomous or semi-autonomous decision-making systems, both of which could use
the detection approaches presented in this paper. The semi-autonomous system use is relevant, because
human-based recognition is highly susceptible to error. Human recognition processes generally require a
huge amount of data processing and it can be a tedious task. Furthermore, in some cases the operator must
be previously trained.
This paper analyzes the applicability of the Haar-like cascade classifier [5], the LBP (Linear Binary
Patterns) cascade classifier [6] and the HOG+SVM (Histogram of Oriented Gradients with Support Vector
1
juliano.cruz@lac.inpe.br
2
elcio@ieav.cta.br
3
guimarae@ieav.cta.br