Reliable Motion Detection
of
Small Targets in Video with
Low
Signal-to-Clutter Ratios
Scott A. Nichols and
R.
Brian Naylor
Sandia National Laboratories, Security Technology Department
P.
0.
Box 5800, Albuquerque, New Mexico 87185-0780
sanicho@sandia.gov (505) 844-1607
rbnaylo@sandia.gov (505) 844-8523
e?
FCEIVED
3UN
19
1995
OSTI
Abstract
Studies show that vigilance decreases rapidly after several minutes when human operators are required to search live
video for infrequent intrusion detections. Therefore, there is a need for systems which can automatically detect
targets in live video and reserve the operator's attention for assessment only. Thus far, automated systems have not
simultaneously provided adequate detection sensitivity, false alarm suppression, and
ease
of setup when used in
external, unconstrained environments. This unsatisfactory performance can be exacerbated by poor video imagery
with low contrast, high noise, dynamic clutter, image misregistration, and/or the presence of small, slow, or
erratically moving targets.
This paper describes
a
highly adaptive video motion detection and tracking algorithm which has been developed
as
part of Sandia's Advanced Exterior Sensor
(AES)
program. The
AES
is a wide-area detection and assessment
system for use in unconstrained exterior security applications. The
AES
detection and tracking algorithm provides
good performance under stressing data and environmental conditions. Features of the algorithm include: reliable
detection with negligible false alarm rate of variable velocity targets having low signal-to-clutter ratios; reliable
tracking of targets that exhibit motion that is non-inertial, Le., varies in direction and velocity; automatic adaptation
to both infrared and visible imagery with variable quality; and suppression of false alarms caused by sensor flaws
andlor cutouts.
NOTE:
This work was supported by the Defense Nuclear Agency and the Department of Energy under contract
DE-AC04-94AJX5000.
Keywords:
video motion detection, clutter suppression, adaptation, false alarm suppression, detection and tracking,
human and vehicle tracking
1.
Introduction
Passive sensors capable of wide-area, stand-off intrusion detection are gaining increased importance in applications
ranging from upgraded fixed perimeter security to rapid-deployment force protection on peace-keeping missions.
Using video motion detection (VMD) technologies increases the system's capabilities by providing not only
intrusion detection, but also providing the operator immediate visual alarm assessment.
VMD systems have been applied to visible imaging and thermal imaging video sensors around perimeters and sensor
fields. However, recent evaluations show nuisance alarms are still excessively high (Vigil 1992), and the sensors and
processors are not designed into an integrated unit. Previous attempts to integrate visible and thermal imaging
devices with advanced processing have produced a significant technology base (Pritchard
1990;
Arlowe
1990)
but a
challenging problem remains: how to provide robust detection using sophisticated processing in an affordable,
reliable package.
Sandia National Laboratories has been tasked by the Defense Nuclear Agency (DNA) to research requirements and
technology for, and develop
a
prototype of, an Advanced Exterior Sensor
(AES)
for ground-based, stand-off
%
biS7RtBUTlON
OF
THIS
DOCUMENT
IS
UNLIMITED
STER
DISCLAIMER
This report was prepared as
an
account
of
work sponsored
by an agency
of
the United States Government. Neither the
United
States
Government nor any agency thereof, nor any
of
their employees, make any warranty, express or implied,
or
assumes any legal liability
or
responsibility
for
the
accuracy, completeness,
or
usefulness
of
any information,
apparatus, product,
or
process disclosed, or represents that
its
use would not infringe privately owned rights. Reference
herein
to
any specific commercial product, process, or
service
by trade name, trademark, manufacturer, or
otherwise does not necessarily constitute
or
imply
its
endorsement, recommendation,
or
favoring
by the United
States
Government or any agency thereof. The views and
opinions
of
authors expressed herein do not necessarily
state
or
reflect
those
of
the United
States
Government or
any agency thereof.
,
DISCLAIMER
Portions
of
this
document may be illegible
in
electronic image products. Images are
produced from the best available original
document.
I
,
wilkirun
0.6x1.65 m
1.0
m2
Crawling
human
head-on
0.5x0.3 m
0.15
m2
intrusion detection and assessment. The
AES
is a 360-degree scanning, multispectral intrusion detection sensor that
simultaneously acquires and processes panoramic images from the infrared, visible, and radar spectrum (Pritchard
1995).
The
AES
is designed to detect humans and vehicles at ranges from 50-1500 meters and moving
as
slowly
as
0.25
metedsecond (0.1 meters/second desired). The current human and vehicle detection requirements for the
AES
are
summarized in Table
1.
visibility
Light rain,
humid
Clear, good
visibility
Light rain,
humid
Table
1.
AES
Detection Range Requirements
I
Conditions
Target
I
Upright human
I
Clear, good
TrucWvan Clear, good
1.5x1.5 m visibility
2.3 m2
Range
500
m
350
m
250
m
200 m
1000 m
Range
(desired)
750 m
525 m
375 m
300 m
1500 m
Other
AES
system goals include: reduction of nuisance alarms by distinguishing humans and vehicles from other
moving objects (birds, tree limbs, debris), classification of at least 90 percent of potential threats versus non-threats,
and continuous detection while assessing alarms from other locations.
To accomplish
as
many of the
AES
detection goals
as
possible, a robust detection and tracking algorithm is required
that can reduce false alarms caused by background and sensor noise; suppress high contrast edges and boundaries in
the imagery; tolerate slight sensor misregistrations; reliably detect small, slow moving, or erratic targets;
automatically adapt to various qualities of imagery and environmental conditions; and reliably detect targets with low
SCR and high electrical, spatial, and temporal noise conditions.
2.
Technical Approach
Development of the detection and tracking algorithm for the
AES
system began in February 1992. Several literature
searches were undertaken to identify the latest published
VMD
technologies in these subject areas: image
registration, clutter suppression, threshold setting and adaptation, target segmentation, data association, target
tracking, and sensor fusion (Pritchard and Nichols 1992).
Sandia’s goal was not to invent new technologies to increase detection performance. Rather, it
was
to identify
promising technologies which were nearly mature, or technologies which could be integrated into an operational
system with a minimum of additional research and development. Technologies that may have promised good
performance but were very new, untested, too computationally intensive, or constrained in application scope were
not pursued.
The review concluded that, in general, 3-D spatio-temporal matched filters with track-before-detect algorithms or
coherent integration algorithms would provide the best detection performance. However, they were deemed
unsuitable for
AES
applications because statistical information about the targets and background
was
incomplete or
unknown, and because they had excessive computational requirements in the temporal image domain. However, the
authors concluded that judicious enhancement of more typical non-integrating clutter suppression, segmentation,
clustering, tracking, and sensor data fusion technologies could still satisfy the
AES
detection requirements.
A
high-
level block diagram
of
the resultant processing architecture is shown in Figure
1.
Data
Pipelines
b
----------
Control
Adaptation
Processing
Figure
1.
AES
Detection and Tracking System Block Diagram
The two shaded blocks in the figure illustrate the principal modules of the infrared and visible
AES
detection and
tracking algorithm pipelines-the topics of this paper. The radar processing pipeline looks much like the infrared
and visible pipelines and follows similar processing steps; however, the resolution and nature of the
AES
radar data
calls for a somewhat different processing approach. For this reason its details
are
outside the scope of this paper.
The other blocks, such
as
data fusion and target analysis, have not been implemented at the time of this report.
Over the course of the advanced algorithm development, four areas have been identified
as
the keys that provided the
best tradeoff between performance and increased memory and computation requirements. These areas are:
1.
Advanced clutter suppression
2.
Multilayered thresholding and feedback adaptation
3.
Purposeful motion analysis
4.
Target prediction and expectation
Advanced Clutter Suppression
The
AES
detection and tracking algorithm uses a unique approach to remove
the
background clutter and noise which
can result in spurious detections not caused by coherent target motion. This method is based on a computationally
efficient approximate solution to the spatio-temporal constraint equation (STCE) commonly used
to
compute image
flow fields.
A
detailed derivation and discussion of this equation is provided in previous literature, hence, only the
final equation is presented here (Horn and Schunk
1980).