QUADRATIC MODELS FOR CURVED LINE DETECTION IN SAR CCD
Davis E. King and Rhonda D. Phillips
MIT Lincoln Laboratory
{davis.king,rhonda.phillips}@ll.mit.edu
ABSTRACT
Synthetic Aperture Radar Coherent Change Detection (SAR CCD) is a sensitive change detector capable of finding ground
surface height changes on the order of a radar wavelength. SAR CCD also has a high false alarm rate, with many false alarms
caused by vegetation that is easily displaced by wind, rain, etc. This work introduces a robust method for detecting long curving
structures in this challenging medium. To demonstrate the accuracy of the method, we test it on the task of automatically
distinguishing curving tire tracks from background clutter.
Index Terms— SAR, coherent change detection, image filtering
1. INTRODUCTION
Synthetic Aperture Radar Coherent Change Detection (SAR CCD) is a sensitive change detector that reveals scene changes
that are on the order of radar wavelength rather than range resolution. SAR CCD has applications in activity monitoring (by
detecting track imprints) [1] [2], search and rescue operations [3], and more. The detection of vehicle tracks is possible because
vehicles make small changes on the ground that slightly alter SAR magnitude and phase values. The automatic detection of
these curvilinear tracks is difficult because the coherence values on such a track are not uniformly low. Furthermore, there are
many naturally occurring false alarms (low coherence that is not the result of scene change) throughout the image that confuse
line detectors. In this paper, we introduce a curvilinear line detection method designed specifically for SAR CCD images. The
method uses two steps, one to detect curve points, and one to join points together to form curves. The first step accounts for
the nonuniform coherence values on line segments, and the second step is necessary to eliminate false alarms. The following
section describes SAR CCD processing and elaborates on the difficulty in detecting line segments in SAR CCD. Section 3
describes the curve detection algorithm, and Section 4 presents experimental results.
2. SAR CCD
SAR CCD is formed using two registered SAR images collected from identical geometries. The collection geometry require-
ment is necessary because SAR CCD exploits differences in phase to detect subtle disturbances on the ground. [4] and [5]
provide a comprehensive overview of the processing required to perform coherent change detection. CCD images are coher-
ence maps between two SAR images, f and g, where the coherence value at each pixel, γ, represents the coherence between
that complex pixel in f and g. CCD is actually an estimated coherence, ˆγ, which is a random variable that depends on the true
coherence and the number of samples used to compute the estimate [6]. The random nature of ˆγ and of the noise present in f
This work is sponsored by the Department of Defense under Air Force Contract FA8721-05-C-0002. Opinions, interpretations, conclusions, and recom-
mendations are those of the authors and are not necessarily endorsed by the United States Government.
