Fringe-adjusted joint transform correlator based
target detection and tracking in forward
looking infrared image sequence
Mohammad S. Alam, FELLOW SPIE
M. Haque
J. F. Khan
H. Kettani
University of South Alabama
Department of Electrical and Computer
Engineering
Mobile, Alabama 36688
E-mail: malam@usouthal.edu
Abstract. The joint transform correlator (JTC) technique has been
found to be suitable for real time pattern recognition applications. Among
the various JTC techniques proposed in the literature, the fringe-
adjusted JTC has been found to yield better correlation output for target
detection. We propose a generalized fringe-adjusted JTC (GFJTC)
based algorithm for efficient detection and tracking of a target in a video
sequence. The proposed algorithm has been found to be suitable for
near real-time recognition and tracking of a static or moving target, while
accommodating the detrimental effects of background variations as
well as other artifacts. The performance of the proposed technique has
been verified with real life forward looking infrared (FLIR) image
sequences.
©
2004 Society of Photo-Optical Instrumentation Engineers.
[DOI: 10.1117/1.1731236]
Subject terms: target detection; joint transform correlation; fringe-adjusted filter;
joint power spectrum; forward-looking infrared image sequence.
Paper 030503 received Oct. 9, 2003; revised manuscript received Dec. 1, 2003;
accepted for publication Dec. 21, 2003.
1 Introduction
The joint transform correlator 共JTC兲
1
technique has shown
remarkable promise for real time optical pattern recognition
and tracking applications.
2,3
The main advantages of the
JTC are that it does not require any complex filter fabrica-
tion, allows real time update of the reference image, per-
mits simultaneous Fourier transform of the reference image
and the unknown input scene, and avoids the precise posi-
tioning otherwise required in matched filter based
correlators.
4
A new class of JTCs that employs Fourier plane joint
power spectrum 共JPS兲 apodization based on the reference
image has been found to yield better correlation output.
Among the various apodization based techniques, the pro-
posed fringe-adjusted filter 共FAF兲 based JTC appears to be
particularly attractive, since it avoids the problems other-
wise associated with alternate JTC techniques.
1–3,5–7
The
fringe-adjusted JTC
8
employs a real valued filter called
FAF at the Fourier plane, and the JPS is multiplied by FAF
before applying the inverse Fourier transform to yield the
correlation output.
9
The fringe-adjusted JTC has been
found to yield better correlation performance than alternate
JTCs
1–3,6
for the case of noise free single target and multi-
target binary input scenes under normal as well as poor
illumination condition.
8–12
However, for noise corrupted
input scenes, whenever the reference power spectrum con-
tains very low values, a fringe-adjusted JTC has been found
to yield low correlation peak intensity. To overcome this
limitation, recently a fractional fringe-adjusted JTC
13
tech-
nique has been proposed that employs a family of real val-
ued filters, called generalized fringe-adjusted filters
共GFAFs兲.
Although various JTC techniques have been employed
for detecting target共s兲 from unknown input scenes, very
little work has been reported in the literature for target
tracking in real-life scenarios. Tam et al.
14
proposed a clas-
sical JTC based target tracking technique
14
using sample
binary images for a limited number of frames. However, in
a real life scenario, target tracking is more challenging be-
cause the input scene may contain target like objects; ob-
jects much brighter than the target; partially occluded tar-
gets; blurry targets; targets blending with the background;
and targets affected by rotation and scale variations as well
as other 3-D distortions. To overcome the aforementioned
limitations, it is necessary to develop a more robust and
efficient target tracking algorithm.
In matched spatial filter-based tracking,
15
the reference
image is generated from several hundred training images
where each training image represents a unique appearance
of the desired target with respect to size and/or in-plane
and/or out-of-plane rotation and scale variations. We dis-
cuss the development and implementation of an algorithm
based on GFJTC, capable of detecting and tracking a target
in gray-level video sequences acquired by an infrared cam-
era mounted on a moving platform. In our case, the refer-
ence is a single image that is updated adaptively from one
frame to the next frame of a sequence. Tracking is achieved
by applying GFJTC by correlating consecutive frames of a
sequence and additional postprocessing of the correlation
output. The performance of the proposed technique has
been verified with real-life forward looking infrared 共FLIR兲
video sequences.
1407Opt. Eng. 43(6) 1407–1413 (June 2004) 0091-3286/2004/$15.00 © 2004 Society of Photo-Optical Instrumentation Engineers