COMPUTER VISION,GRAPHICS, AND IMAGE PROCESSING ]I, 280-293 (1983)
NOTE
Application of the One-Dimensional Fourier
Transform for Tracking Moving Objects
in Noisy Environments*
SARAHA. RAJALA, ALFY N. RIDDLE, AND WESLEY E. SNYDER
North Carolina State University, Electrical Engineering Department, Box 52 75, Raleigh, North
Carolina 2 7650
Received July 20, 1982
In Riddle and Rajala (Fifteenth Asilomar Conference on Circuits, Systems and Computers,
November 1981), an algorithm was presented which operates on an image sequence to identify
all sets of pixels having the same velocity. The algorithm operates by performing a transforma-
tion in which all pixels with the same two-dimensional velocity map to a peak in a transform
space. The transform can be decomposed into applications of the one-dimensional Fourier
transform and therefore can gain from the computational advantages of the FFT. The aim of
this paper is the concern with the fundamental limitations of that algorithm, particularly as
relates to its sensitivity to image-disturbing parameters as noise, jitter, and clutter, A modifica-
tion to the algorithm is then proposed which increases its robustness in the presence of these
disturbances.
1. INTRODUCTION
This paper contains the development of an algorithm for identifying the tracks of
moving objects in noisy environments. An example application is a sensor on a
downward-looking space platform whose objective is to detect moving aircraft. Even
if the background can be assumed constant, simple differencing techniques are
inadequate to detect a low amplitude target in the presence of camera jitter or clutter
(time-varying spatially correlated noise).
A technique is presented in this paper which characterizes a target uniquely by its
motion. That motion, over time, creates a track. These tracks, as well as the objects
creating the tracks, can be identified using an algorithm which is based on the
application of the one-dimensional Fourier transform. The result is an ability to
simultaneously acquire the tracks of, and identify, resolved and unresolved targets in
the presence of noise.
2. BACKGROUND
Sophisticated algorithms for tracking multiple targets in environments with addi-
tive noise exist, and a good review of these techniques is presented in [2]. These
algorithms basically fall into two categories, non-Bayesian and Bayesian, and are ~/11
used in active tracking systems (i.e., radar). Both approaches, although quite robust,
require a large number of computations for each new update. An alternative
approach [3] combines successive sets of two-dimensional frames and then screens
these using a 3 × 3 pixel window. The result is a simple, computationally efficient
algorithm. This algorithm, however, requires two major assumptions: (1) the target
*This work was sponsored in part by NASA Langley Research Center Grant NAG-I-20.
280
0734-189X/83/020280-14503.00/0
Copyright © 1983 by Aettdernie Press. Inc.
All rights of reproductionin any formreserved.