IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 13, NO. 6, JUNE 2016 801
Detecting Cars in VHR SAR Images
via Semantic CFAR Algorithm
Yong Huang and Fang Liu, Senior Member, IEEE
Abstract—In this letter, a novel semantic constant-false-
alarm-rate (CFAR) method for the detection of cars from a
very high resolution synthetic-aperture-radar (SAR) image is pre-
sented. The method not only employs the strong scattering features
of the target which is used in CFAR but also employs the shadow
features of the target. Furthermore, the semantic relationship
between the strong scattering features and the shadow features
is established to partly reduce the false alarm targets. By the
experiments on the MiniSAR image of 4-in resolution, it is shown
that the proposed semantic CFAR method outperforms the CFAR
algorithm by a much lower false alarm rate.
Index Terms—Constant false alarm rate (CFAR), semantic,
synthetic aperture radar (SAR), target detection, very high
resolution (VHR).
I. INTRODUCTION
I
N THE last several years, urban areas have been charac-
terized by air pollution and traffic jams. A lot of local
municipalities have directed their atten tion to the problems of
preventing traffic jams and limiting automobile exh aust pollu-
tion, which connected to estimate the number of cars by the car
detection technique. A great number of video cameras are set to
monitor the concentration of vehicles in many cities, but they
are vulnerable to rain, snow, and fog. The synthetic aperture
radar (SAR) technique captured great research interests because
it can be applied on all weather while the optical monitor
depends on the sunshine and weather conditions.
In the curre nt literature, several car detection techniques are
mainly based on optical imagery. Leitloff et al. proposed an
approach using Haar-like f eatures to detect cars from optical
satellite images. The single cars were identified by means of
Manuscript received December 13, 2015; re vised January 21, 2016 and
February 21, 2016; accepted March 11, 2016. Date of publication April 13,
2016; date of current version May 19, 2016. This work was supported in part
by the National Basic Research Program (973 Program) of China under Grant
2013CB329402, by the National Natural Science Foundation of China under
Grants 61573267, 61571342, and 61572383, by the Program for Cheung Kong
Scholars and Innovative Research Team in University under Grant IRT_15R53,
by the Major Research Plan of the National Natural Science Foundation of
China under Grants 91438201 and 91438103, and by the Fund for Foreign
Scholars in Univ ersity Research and Teaching Programs (the 111 Project) under
Grant B07048.
The authors are with the School of Computer Science and Technology and
also with the Key Laboratory of Intelligent Perception and Image Understand-
ing of the Ministry of Education, International Research Center for Intelligent
Perception and Computation, Joint International Research Laboratory of Intel-
ligent Perception and Computation, Xidian University, Xi’an 710071, China
(e-mail: chinahe0609@aliyun.com; f63liu@163.com).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/LGRS.2016.2546309
Fig. 1. Diagram of the block scheme of the car detection in VHR SAR images.
line extraction technique [1]. Salehi et al. presented a three-step
framework for the automatic extraction of moving vehicles.
The method employed light-detection-and-ranging systems to
detect a vehicle in the panchromatic and mu ltispectral images
[2]. Yao and Stilla extracted vehicles by a grid cell, an d 3-D
point-cloud-analysis-based methods were compared with other
methods [3]. Moranduzzo and Melgani proposed a car-
detecting method which computes a similarity measure with
a catalog of cars used as reference [4]. Bai et al. [5] and
Zhang et al. [6] presented an object detection method based on
structural feature extraction and selection.
The constant false alarm ra te (CFAR) is an important method
for target detection in a SAR image. In the past, the CFAR
algorithm was mainly used to detect the big target because the
resolution of the SAR image was low [7]–[9]. Nowadays, with
the rapid increase of the resolution of the SAR image, the CFAR
method was used to detect the small targets in a high-resolution
SAR image [10], [11]. At present, very high spatial resolution
airborne SAR acquiring data with decimeter resolutions has
become widely available. A veryhighresolution(VHR)SAR
image provides a lot of details of the object, so we can describe
the structural features of the target which appeared as a few
pixels in the past. The 0.1-m spatial resolution SAR imaging
system has been achieved by the MiniSAR system, the PAMIR
system, and the SETHI system [12]–[14]. These data of VHR
SAR images h ave the potential to be employed for detecting
various important objects, such as cars, aircrafts, buildings, and
other objects.
The CFAR algorithm is based on the modeling of the statisti-
cal distribution of local background clutter measurements. The
1545-598X © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.