To assist the warfighter in visually identifying potentially dangerous roadside objects, the U.S. Army RDECOM
CERDEC Night Vision and Electronic Sensors Directorate (NVESD) has developed an elevated video sensor system
testbed for data collection. This system provides color and mid-wave infrared (MWIR) imagery. Signal Innovations
Group (SIG) has developed an automated processing capability that detects the road within the sensor field of view and
identifies potentially threatening buried objects within the detected road. The road detection algorithm leverages system
metadata to project the collected imagery onto a flat ground plane, allowing for more accurate detection of the road as
well as the direct specification of realistic physical constraints in the shape of the detected road. Once the road has been
detected in an image frame, a buried object detection algorithm is applied to search for threatening objects within the
detected road space. The buried object detection algorithm leverages textural and pixel intensity-based features to detect
potential anomalies and then classifies them as threatening or non-threatening objects. Both the road detection and the
buried object detection algorithms have been developed to facilitate their implementation in real-time in the NVESD
system.
Ground penetrating radar (GPR) and electromagnetic induction (EMI) sensors provide complementary capabilities in
detecting buried targets such as landmines, suggesting that the fusion of GPR and EMI modalities may provide
improved detection performance over that obtained using only a single modality. This paper considers both pre-screening
and the discrimination of landmines from non-landmine objects using real landmine data collected from a
U.S. government test site as part of the Autonomous Mine Detection System (AMDS) landmine program. GPR and
EMI pre-screeners are first reviewed and then a fusion pre-screener is presented that combines the GPR and EMI prescreeners
using a distance-based likelihood ratio test (DLRT) classifier to produce a fused confidence for each pre-screener
alarm. The fused pre-screener is demonstrated to provide substantially improved performance over the
individual GPR and EMI pre-screeners.
The discrimination of landmines from non-landmine objects using feature-based classifiers is also considered. The
GPR feature utilized is a pre-processed, spatially filtered normalized energy metric. Features used for the EMI sensor
include model-based features generated from the AETC model and a dipole model as well as features from a matched
subspace detector. The EMI and GPR features are then fused using a random forest classifier. The fused classifier
performance is superior to the performance of classifiers using GPR or EMI features alone, again indicating that
performance improvements may be obtained through the fusion of GPR and EMI sensors. The performance
improvements obtained both for pre-screening and for discrimination have been verified by blind test results scored by
an independent U.S. government contractor.
Previous work has introduced a framework for information-based sensor management that is capable of tasking multiple
sensors searching for targets among a set of discrete objects or in a cell grid. However, in many real-world scenarios--
such as detecting landmines along a lane or road--an unknown number of targets are present in a continuous spatial
region of interest. Consequently, this paper introduces a grid-free sensor management approach that allows multiple
sensors to be managed in a sequential search for targets in a grid-free spatial region. Simple yet expressive Gaussian
target models are introduced to model the spatial target responses that are observed by the sensors. The sensor manager
is then formulated using a Bayesian approach, and sensors are directed to make new observations that maximize the
expected information gain between the posterior density on the target parameters after a new observation and the current
posterior target parameter density. The grid-free sensor manager is applied to a set of real landmine detection data
collected with ground penetrating radar (GPR) and electromagnetic induction (EMI) sensors at a U.S. government test
site. Results are presented that compare the performance of the sensor manager with the performance of an unmanaged
joint pre-screener that fuses individual GPR and EMI pre-screeners. The sensor manager is demonstrated to provide
improved detection performance while requiring substantially fewer sensor observations than are made with the
unmanaged joint pre-screening approach.
KEYWORDS: Sensors, Magnetometers, Data modeling, Magnetic sensors, Sensor performance, Target detection, Binary data, Land mines, Electromagnetic coupling, Visual process modeling
In previous work, a sensor management framework has been developed that manages a suite of sensors in a search for
static targets within a grid of cells. This framework has been studied for binary, non-binary, and correlated sensor
observations, and the sensor manager was found to outperform a direct search technique with each of these different
types of observations. Uncertainty modeling for both binary and non-binary observations has also been studied. In this
paper, a new observation model is introduced that is motivated by the physics of static target detection problems such as
landmine detection and unexploded ordnance (UXO) discrimination. The new observation model naturally
accommodates correlated sensor observations and models both the correlation that occurs between observations made
by different sensors and the correlation that occurs between observations made by the same sensor. Uncertainty
modeling is also implicitly incorporated into the observation model because the underlying parameters of the target and
clutter cells are allowed to vary and are not assumed to be constant across target cells and across clutter cells. Sensor
management is then performed by maximizing the expected information gain that is made with each new sensor
observation. The performance of the sensor manager is examined through performance evaluation with real data from
the UXO discrimination application. It is demonstrated that the sensor manager is able to provide comparable detection
performance to a direct search strategy using fewer sensor observations than direct search. It is also demonstrated that
the sensor manager is able to ignore features that are uninformative to the discrimination problem.
Previous research has developed an information-theoretic sensor management framework for improving static target
detection performance. This framework has been successfully applied to a large dataset of real landmine data;
performance using the sensor manager on this dataset was demonstrated to be superior to performance using a direct
search technique in which sensors blindly sweep through the gridded region of interest. In previous work, the sensor
manager has modeled the observations made in each grid cell as being independent from the other observations made in
that cell by the same sensor and also as being independent from observations made in that cell by other sensors. Such a
modeling approach fails to account for the correlations that will result between observations made both by the same and
different sensors. This paper alters the modeling framework that has been used previously to incorporate observation
correlation, which will more realistically model the interrelationships between sensor observations. After introducing
the new modeling approach, results are then presented that compare the performance of the sensor manager to the
performance of an unmanaged direct search procedure. The sensor manager is again demonstrated to outperform direct
search. Furthermore, the performance effects of modeling and failing to model correlation are examined through
simulation. Failing to model correlation that is present in the data is demonstrated to substantially degrade performance
and cause direct search to outperform the sensor manager. However, when correlated modeling is used to model
correlated data, the sensor manager is again demonstrated to outperform direct search.
Previous work by the authors using information-based sensor management for static target detection has utilized a
probability of error performance metric that assumes knowledge of the number of targets present in a grid of cells.
Using this probability of error performance metric, target locations are estimated as the N cells with the largest posterior
state probabilities of containing a target. In a realistic application, however, the number of targets is not known a priori.
The sequential probability ratio test (SPRT) developed by Wald is therefore implemented within the previously
developed sensor management framework to allow cell-level decisions of "target" or "no target" to be made based on
the observed sensor data. Using these cell-level decisions, more traditional performance metrics such as probability of
detection and probability of false alarm may then be calculated for the entire region of interest.
The resulting sensor management framework is implemented on a large set of data from the U.S. Army's autonomous
mine detection sensors (AMDS) program that has been collected using both ground penetrating radar (GPR) and
electromagnetic induction (EMI) sensors. The performance of the sensor manager is compared to two different direct
search techniques, and the sensor manager is found to achieve the same Pd performance at a lower cost than either of the direct search techniques. Furthermore, uncertainty in the sensor performance characteristics is also modeled, and the
use of uncertainty modeling allows a higher Pd to be obtained than is possible when uncertainty is not modeled within the sensor management framework.
A proliferation of the number and variety of sensors for the landmine detection problem has created the need for a sensor manager that is able to intelligently task and coordinate the operation of a suite of landmine sensors. Previous work has developed a framework for sensor management that takes into account the context of the landmine detection problem. The sensor manager searches for N targets in a grid using M multimodal sensors by seeking to maximize the expected information gain. The probabilities of detection and false alarm of the sensors are assumed to be known and are used in the sensor manager calculations. However, in a real-world landmine detection setting, the performance characteristics of the sensors will in fact be unknown. Uneven and irregular ground, vegetation, unanticipated clutter objects, even bad weather - all these can affect the performance of a landmine sensor. This paper examines the effects of uncertainty in the probabilities of detection and false alarm on the performance of the previously presented sensor manager and further examines the performance effects of properly and improperly modeling this uncertainty. Performance is, naturally, found to be adversely affected by uncertainty. However, it is demonstrated that properly modeling the uncertainty present in the problem helps to recover some of the performance that is lost through the introduction of uncertainty.
KEYWORDS: Sensors, Land mines, Sensor performance, Active sensors, Target detection, Detection and tracking algorithms, General packet radio service, Electromagnetic coupling, Binary data, Computer simulations
We consider an information-theoretic approach for sensor management that chooses sensors and sensor parameters in order to maximize the expected discrimination gain associated with each new sensor measurement. We analyze the problem of searching for N targets with M multimodal sensors, where each sensor has its own probability of detection, probability of false alarm, and cost of use. Other information, such as the prior distribution of the targets in space and the degree of constraint of the sensor motion, is also utilized in our formulation. Performance of the sensor management algorithm is then compared to the performance of a direct-search procedure in which the sensors blindly search through all cells in a predetermined path. The information-based sensor manager is found to have significant performance gains over the direct-search approach. Algorithm performance is also analyzed using real landmine data taken with three different sensing modalities. Detection performance using the sensor management algorithm is again found to be superior to detection performance using a blind search procedure. The simulation and real-data results also both illuminate the increased performance available through multimodal sensing.
KEYWORDS: Mining, Land mines, Sensors, Detection and tracking algorithms, Signal to noise ratio, General packet radio service, Algorithm development, Statistical analysis, Optical spheres, Antennas
In this paper we use a sequential probability ratio test (SPRT) on ground penetrating radar (GPR) data to detect buried antipersonnel land mines and to reject clutter objects. Detection is performed for both fixed-depth and variable-depth cases. We use high-dimensional analysis of variance (HANOVA) to window the GPR data before SPRT analysis. Our algorithm uses a library of mine and clutter objects and performs a series of SPRTs using the object library for each unknown image. We also evaluate the performance of our fixed-depth and variable-depth detection algorithms versus noise and SPRT threshold value.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.