AUTOMATED EXTRACTION OF AIRPORT RUNWAY PATTERNS FROM RADAR IMAGERY 
a 
F 
Mr. Richard A. Hevenor, Electronics Engineer and 
Dr. Pi-Fuay Chen, Electronics Engineer 
U.S. Army Topographic Engineering Center 
Fort Belvoir, Virginia 22060-5546 
United States of America 
ISPRS Technical Commission III 
ABSTRACT: 
A method is presented to extract linear terrain features from synthetic aperture radar imagery. An 
input radar image is smoothed with an edge preserving smoothing operation. Edge detection is performed 
using a Sobel operator, and both the magnitude and the directional images are computed. The edges are 
strengthened using several iterations of a relaxation operation in which both the magnitude image and 
the directional image are updated with each iteration. The output of the relaxation operation is a 
binary edge image, which is thinned. A connected components routine is run in which two passes through 
the image are used to provide a unique label for each connected component. The connected components 
related only to the runway pattern are extracted by computing certain properties of each component. A 
border-following algorithm is used to follow only the outermost borders and give each of the pixels on 
an outermost border a maximum brightness value. A tracking algorithm is used to change the binary 
image array into a set of Freeman chain codes, which serve as the input to a line-forming routine that 
uses a standard polygon approximation algorithm. Experimental results on a real synthetic aperture 
radar image are presented. 
KEY WORDS: Image Analysis, Image Interpretation, Image Processing, Machine Vision, SAR 
Fi 
INTRODUCTION applied that begin with edge preserving smoothing : 
and end with polygon approximation. Some of the 
The problem of extracting airport runway patterns processing routines are elementary, such as edge 
from optical photography has been the subject of detection; however, some of them are very com- 
study for several years (Nevatia and Babu, 1980). plicated, such as relaxation for edge reinforce- 
In the referenced work, the emphasis was on low ment and border following. The purpose of edge 
level vision computations, and little effort was preserving smoothing is to eliminate noise and at 
made to isolate the connected components of the the same time preserve edges so that they are not 
airfield. Also, very little work has been done blurred. The edge detection algorithm will en- 
in extracting airport runway patterns from syn- hance edges and compute the direction of the edge 
thetic aperture radar imagery. Since airports at each pixel. The direction of each edge is 
represent a potential military target, they needed as an input to the next algorithm which is 
should be extracted from radar imagery as quickly relaxation for edge reinforcement. The purpose 
as possible. of this routine is to enhance edges using the 
contextual information of the surrounding neigh- 
The purpose of this paper is to present a syste- borhood. The output of the relaxation calcula- ; 
matic procedure consisting of a number of image tions is a binary edge image in which many of the Fi; 
processing algorithms that allow one to go from strong edges are too thick. A thinning operation 
an original radar image containing an airfield to is required in order to thin most edges down to 
a binary image consisting only of components that the thickness of 1 pixel in width. The connected Ed; 
are related to the airfield. All connected com- components algorithm provides a unique label (num- 
ponents in the image that relate to other terrain ber) for each pixel in a given component of The 
features will be eliminated. No attempt was made l-pixels. Use is made of the definition of 8-con- rit 
to make the procedure robust or general. The nectivity for l-pixels and 4-connectivity for frc 
main objective was to see if the components of O-pixels. These definitions were also used for in 
an airfield could be isolated for the sample the thinning operation. Eleven region property det 
image used. This image was a 512 by 512 pixel calculations were performed to isolate the con- Smc 
image that contained an airfield located near nected components that belong only to the airfield The 
Elizabeth City, North Carolina (see Figure 1). A border-following algorithm is used to determine ing 
The radar system used to obtain this image was and uniquely label all of the l-pixels that exist the 
the UPD-4 system which is an X-band radar with between a given connected component of l-pixels pix 
HH-polarization. The radar image was digitized and a connected component of O-pixels. Two bor- are 
with 8-bits. The image processing algorithms der-following algorithms were implemented for this cen 
were written in the LISP programming language work. The first algorithm finds and labels all pen 
and executed on a Symbolics 3670 LISP machine. border pixels for outer borders and hole borders. hex 
The same algorithms were later recoded in the C The second algorithm finds just the outermost bor- squ 
programming language and implemented on a SUN ders. The difference between outer borders and for 
3/180 microcomputer system for ease of transfer hole borders will be explained in detail later. out 
to a development laboratory. The rest of this After border-following has been completed, lists of 
paper will discuss the procedure used to extract are generated, each of which consists of an 8-dir- ter 
the airfield from the radar image. Each of the ection chain together with the coordinates of the 5- 
algorithms used in this procedure will be briefly initial point in the chain. These lists are for 
discussed. Finally, the results obtained after referred to as Freeman chain codes and are used as est 
applying all the algorithms will be presented. input data to a polygon approximation routine. of 
The purpose of polygon approximation is to approx- mea: 
METHODOLOGY imate irregular curves with a consecutive series ure 
of line segments. The particular polygon approx- 75 
The procedure for extracting the airfield from imation routine used will be discussed later. The 9 p 
the image shown in Figure 1 is given in Figure 2. following sections of this paper discuss in more 
In this procedure, processing routines are detail each of the algorithms mentioned above. 
808