This PDF file contains the front matter associated with SPIE Proceedings Volume 8393, including the Title Page, Copyright information, Table of Contents, and the Conference Committee listing.

Chemical and biological monitoring systems are faced with the challenge of detecting weak signals from contam- inants of interest while at the same time maintaining extremely low false alarm rates. We present methods to control the number of false alarms while maintaining power to detect; evaluating these methods on a fixed sensor grid. Contaminants are detected using signals produced from underlying sensor-specific detection algorithms. By learning from past data, an adaptive background model is constructed and used with a multi-hypothesis testing method to control the false alarm rate. Detection methods for chemical/biological releases often depend on specific models for release types and missed detection rates at the sensors. This can be problematic in field situations where environment specific effects can alter both a sensor's false alarm and missed detection characteristics. Using field data, the false alarm statistics of a given sensor can be learned and used for inference; however the missed detection statistics for a sensor are not observable while in the field. As a result, we pursue methods that do not rely on accurate estimates of a sensor's missed detection rate. This leads to the development of the Adaptive Regions Method that under certain assumptions is designed to conservatively control the expected rate of false alarms produced by a fusion system over time, while maintaining power to detect.

With recent changes in threats and methods of warfighting and the use of unmanned aircrafts, ISR (Intelligence, Surveillance and Reconnaissance) activities have become critical to the military's efforts to maintain situational awareness and neutralize the enemy's activities. The identification and tracking of dismounts from surveillance video is an important step in this direction. Our approach combines advanced ultra fast registration techniques to identify moving objects with a classification algorithm based on both static and kinematic features of the objects. Our objective was to push the acceptable resolution beyond the capability of industry standard feature extraction methods such as SIFT (Scale Invariant Feature Transform) based features and inspired by it, SURF (Speeded-Up Robust Feature). Both of these methods utilize single frame images. We exploited the temporal component of the video signal to develop kinematic features. Of particular interest were the easily distinguishable frequencies characteristic of bipedal human versus quadrupedal animal motion. We examine limits of performance, frame rates and resolution required for human, animal and vehicles discrimination. A few seconds of video signal with the acceptable frame rate allow us to lower resolution requirements for individual frames as much as by a factor of five, which translates into the corresponding increase of the acceptable standoff distance between the sensor and the object of interest.

We demonstrate that human skin biometrics in the visible to near infrared (VNIR) regime can be used as reliable features in a multistage human target tracking algorithm suite. We collected outdoor VNIR hyperspectral data of human skin, consisting of two human subjects of different skin types in the Fitzpatrick Scale (Type I [Very Fair] and Type III [White to Olive]), standing side by side at seven ranges (50 ft to 370 ft) in a suburban background. At some of these ranges, the subjects fall under the small target category. We propose a three-step approach: Step 1, reflectance retrieval; Step 2, exploitation of absorption wavelength line at 577 nanometers, due to oxygenated hemoglobin in blood near the surface of skin; and Step 3, matched filtering on candidate patches in the input imagery that successfully passed Step 2, using as input all of the available bands in a spectral average representation of human skin. Step-3 functionality is only applied to patches in the imagery showing evidence of human skin (Step 2 output). Regardless of the targets' kinematic states, the approach produced some excellent results locating the presence of human skin in the example dataset, yielding zero false alarms from potential confusers in the scene. The approach is expected to function as the focus of attention stage of a multistage algorithm suite for human target tracking.

The parametric Rao test for multichannel adaptive signal detection by the adaptive generalized detector (GD) constructed based on the generalized approach to signal processing in noise is derived by modeling the disturbance signal as a multichannel autoregressive process. The parametric Rao test takes a form identical to that of parametric GD for space-time adaptive processing in airborne surveillance radar systems and other similar applications. The equivalence offers new insights into the performance and implementation of the GD. Specifically, the Rao/GD is asymptotically (in the case of large samples) a parametric generalized likelihood ratio test generalized detector (GLRT GD) due to an asymptotic equivalentce between the Rao test and the GLRT/GD. The asymptotic distribution of the Rao/GD test statistic is obtained in closed form, which follows an exponential distribution under the null hypothesis (the target return signal is absent) and, respectively, a non-central Chi-squared distribution with two degrees of freedom under the alternative hypothesis (the target return signal is present). The noncentrality parameter of the noncentral Chi-squared distribution is determined by the output signal-to-interference-plus-noise ratio of a temporal whitening filter. Since the asymptotic distribution under the null hypothesis is independent of the unknown parameters, the Rao/GD asymptotically achieves constant false alarm rate (CFAR) GD. Numerical results show that these results are superior in predicting the performance of the parametric adaptive matched filter detector even with moderate data support.

Multiple Input Multiple Output- MIMO Radar is a fast growing research area. This paper will give a brief introduction to the subject as well as derive an image formation scheme. The general problem of radar imaging is to use some physical model for a transmitted signal, and measurements of the signal that is scattered back to a receiver by a scene to attempt to derive information about the scene. The concept of communication involves a message sender, a message receiver, and a channel. The sender sends a message through the channel to the receiver. The receiver attempts to recover the original message. MIMO communication is just communication that involves sending several messages to several recipients. The problem of Multiple Input Multiple Output Radar Imaging is to use the corruption of transmitted messages to try and derive useful information about the environment that the messages traveled through. The extra information gained with MIMO Radar can be used to get rid of false targets, detect moving targets, and create a better resolution image. The plan for this research is to culminate to an in-scene 3-d Image reconstruction algorithm. The model presented provides a context in which to examine this problem.

We present a theory and algorithm for detecting and classifying weak, distributed patterns in network data that provide actionable information with quantiable measures of uncertainty. Our work demonstrates the eectiveness of space-time inference on graphs, robust matrix completion, and second order analysis for the detection of distributed patterns that are not discernible at the level of individual nodes. Motivated by the importance of the problem, we are specically interested in detecting weak patterns in computer networks related to Cyber Situational Awareness. Our focus is on scenarios where the nodes (terminals, routers, servers, etc.) are sensors that provide measurements (of packet rates, user activity, central processing unit usage, etc.) that, when viewed independently, cannot provide a denitive determination of the underlying pattern, but when fused with data from across the network both spatially and temporally, the relevant patterns emerge. The approach is applicable to many types of sensor networks including computer networks, wireless networks, mobile sensor networks, and social networks, as well as in contexts such as databases and disease outbreaks.

We derive five new particle flow algorithms for nonlinear filters based on the small curvature approximation inspired by fluid dynamics. We find it extremely interesting that this physically motivated approximation generalizes two of our previous exact flow algorithms, namely incompressible flow and Gaussian flow. We derive a new algorithm to compute the inverse of the sum of two linear differential operators using a second homotopy, similar to Feynman's perturbation theory for quantum electrodynamics as well as Gromov's h-principle.

Many modern agile sensor systems are capable of being adaptively tasked in response to an evolving environment. This paper describes an algorithm developed in the framework of previous work by Casta~non, Wintenby and Krishnamurthy. The goal is to schedule the time and dwell time for updates of targets under track using a phased array radar. This problem is addressed using Lagrangian relaxation, decoupling the joint optimisation into a series of single target problems. After discretising the single target decision state (i.e., the covariance matrix), these single target problems are solved as Markov decision processes. An example of a method for selecting the state space discretisation is outlined and the results used to generate a closed loop schedule for a set of track states.

Steady-state performance of a tracking filter is traditionally evaluated immediately after a track update. However, there is commonly a further delay (e.g., processing and communications latency) before the tracks can actually be used. We analyze the accuracy of extrapolated target tracks for four tracking filters: Kalman filter with the Singer maneuver model and worst-case correlation time, with piecewise constant white acceleration, and with continuous white acceleration, and the reduced state filter proposed by Mookerjee and Reifler.^{1, 2} Performance evaluation of a tracking filter is significantly simplified by appropriate normalization. For the Kalman filter with the Singer maneuver model, the steady-state RMS error immediately after an update depends on only two dimensionless parameters.³ By assuming a worst case value of target acceleration correlation time, we reduce this to a single parameter without significantly changing the filter performance (within a few percent for air tracking).⁴ With this simplification, we find for all four filters that the RMS errors for the extrapolated state are functions of only two dimensionless parameters. We provide simple analytic approximations in each case.

In this paper we present an approach for tracking in long range radar scenarios. We show that in these scenarios the extended Kalman filter is not desirable as it suffers from major consistency problems, and that particle filters may suffer from a loss of diversity among particles after resampling. This leads to sample impoverishment and the divergence of the filter. In the scenarios studied, this loss of diversity can be attributed to the very low process noise. However, a regularized particle filter and the Gaussian Mixture Sigma-Point Particle Filter are shown to avoid this diversity problem while producing consistent results.

Fusing data together for target tracking is a complex problem. There are two key steps. First, the raw observations must be associated with existing tracks or used to form new tracks. Once the association has been done, then the tracks can be updated and filtered with the new data. The updating and filtering is usually the easier of the two parts and it is the association that can lead to most of the complexity in target tracking. When associating data (either measurements or tracks or both) with existing tracks, the separation between the tracks is critical to how difficult the association decisions will be. If the tracks are widely separated then the association decisions can be relatively easy. On the other hand, when the tracks are closely spaced the association decisions can be very difficult or nearly impossible. When the tracks or measurements are in three dimensions (such as with active sensors) the association can be accomplished in all three dimension thus making an easier distinction of targets that may be very close in two dimensions, but distant in the third dimension. However, when there are only two dimensions (as for passive sensors) observed by a sensor, targets that are widely separated may appear to be very close or even unresolved. In this paper, we will discuss the issues involved with applying the Probabilistic Multi-Hypothesis Tracking (PMHT) algorithm to fusing either measurements or tracks from passive sensors.

This paper presents an approach to attribute estimation incorporating data association ambiguity. In modern tracking systems, time pressures often leave all but the most likely data association alternatives unexplored, possibly producing track inaccuracies. Numerica's Bayesian Network Tracking Database, a key part of its Tracker Adjunct Processor, captures and manages the data association ambiguity for further analysis and possible ambiguity reduction/resolution using subsequent data. Attributes are non-kinematic discrete sample space sensor data. They may be as distinctive as aircraft ID, or as broad as friend or foe. Attribute data may provide improvements to data association by a process known as Attribute Aided Tracking (AAT). Indeed, certain uniquely identifying attributes (e.g. aircraft ID), when continually reported, can be used to define data association (tracks are the collections of observations with the same ID). However, attribute data arriving infrequently, combined with erroneous choices from ambiguous data associations, can produce incorrect attribute and kinematic state estimation. Ambiguous data associations define the tracks that are entangled with each other. Attribute data observed on an entangled track then modify the attribute estimates on all tracks entangled with it. For example, if a red track and a blue track pass through a region of data association ambiguity, these tracks become entangled. Later red observations on one entangled track make the other track more blue, and reduce the data association ambiguity. Methods for this analysis have been derived and implemented for efficient forward filtering and forensic analysis.

Using the Automatic Identification System (AIS) ships identify themselves intermittently by broadcasting their location information. However, traditionally radars are used as the primary source of surveillance and AIS is considered as a supplement with a little interaction between these data sets. The data from AIS is much more accurate than radar data with practically no false alarms. But unlike the radar data, the AIS measurements arrive unpredictably, depending on the type and behavior of a ship. The AIS data includes target IDs that can be associated to initialized tracks. In multitarget maritime surveillance environment, for some targets the revisit interval form the AIS could be very large. In addition, the revisit intervals for various targets can be different. In this paper, we proposed a joint probabilistic data association based tracking algorithm that addresses the aforementioned issues to fuse the radar measurements with AIS data. Multiple AIS IDs are assigned to a track, with probabilities updated by both AIS and radar measurements to resolve the ambiguity in the AIS ID source. Experimental results based on simulated data demonstrate the performance the proposed technique.

The Maximum Likelihood Probabilistic Data Association (ML-PDA) tracker is an algorithm that has been shown to work well against low-SNR targets in an active multistatic framework with multiple transmitters and multiple receivers. In this framework, measurements are usually received in time-bearing space. Prior work on ML-PDA implemented the algorithm in Cartesian measurement space - this involved converting the measurements and their associated covariances to (x, y) coordinates. The assumption was made that Gaussian measurement error distributions in time-bearing space could be reasonably approximated by transformed Gaussian error distributions in Cartesian space. However, for data with large measurement azimuthal uncertainties, this becomes a poor assumption. This work compares results from a previous study that applied ML-PDA in a Cartesian implementation to the Metron 2009 simulated dataset against ML-PDA applied to the same dataset but with the algorithm implemented in time-bearing space. In addition to the Metron dataset, a multistatic Monte Carlo simulator is used to create data with properties similar to that in the Metron dataset to statistically quantify the performance difference of ML-PDA operating in Cartesian measurement space against that of ML-PDA operating in time-bearing space.

Effective multi-sensor, multi-target, distributed composite tracking requires the management of limited network bandwidth. In this paper we derive from first principles a value of information for measurements that can be used to sort the measurements in order from most to least valuable. We show the information metric must account for the models and filters used by the composite tracking system. We describe how this value of information can be used to optimize bandwidth utilization and illustrate its effectiveness using simulations that involve lossy and latent network models.

Target tracking techniques have usually been applied to physical systems via radar, sonar or imaging modalities. But the same techniques - filtering, association, classification, track management - can be applied to nontraditional data such as one might find in other fields such as economics, business and national defense. In this paper we explore a particular data set. The measurements are time series collected at various sites; but other than that little is known about it. We shall refer to as the data as representing the Megawatt hour (MWH) output of various power plants located in Afghanistan. We pose such questions as: 1. Which power plants seem to have a common model? 2. Do any power plants change their models with time? 3. Can power plant behavior be predicted, and if so, how far to the future? 4. Are some of the power plants stochastically linked? That is, do we observed a lack of power demand at one power plant as implying a surfeit of demand elsewhere? The observations seem well modeled as hidden Markov. This HMM modeling is compared to other approaches; and tests are continued to other (albeit self-generated) data sets with similar characteristics. Keywords: Time-series analysis, hidden Markov models, statistical similarity, clustering weighted

Target tracking sensors and algorithms are usually evaluated using Monte Carlo simulations covering a large parameter space. We show a tracker for which the evaluation can be greatly simplified. We apply it to the one dimensional crossing track problem (e.g. ground target tracking in a dense target environment, where targets are confined to a road), and estimate the probability that measurements and tracks are incorrectly associated. If only position is measured, we find the probability of a misassociation is a very simple analytic function of the relevant parameters: measurement standard deviation, measurement interval, target density, and target acceleration. For normally distributed target velocities, the average time between misassociations also has a simple form. We suggest roll-up metrics for tracking sensors and tracking problems.

This paper presents algorithms for prediction, tracking, and retrodiction for targets whose motion is constrained by external conditions (e.g., shipping lanes, roads). The targets are moving along a path, defined by way-points and segments. Measurements are obtained by sensors at low revisit rates (e.g., spaceborne). Existing tracking algorithms assume that the targets follow the same motion model between successive measurements, but in a low revisit rate scenario targets may change the motion model between successive measurements. The proposed prediction algorithm addresses this issue by considering possible motion model whenever targets move to a different segment. Further, when a target approaches a junction, it has the possibility to travel into one of the multiple segments connected to that junction. To predict the probable locations, multiple hypotheses for segments are introduced and a probability is calculated for each segment hypothesis. When measurements become available, segment hypothesis probability is updated based on a combined mode likelihood and a sequential probability ratio test is carried out to reject the hypotheses. Retrodiction for path constrained targets is also considered, because in some scenarios it is desirable to find out the target's exact location at some previous time (e.g., at the time of an oil leakage). A retrodiction algorithm is also developed for path constrained targets so as to facilitate motion forensic analysis. Simulation results are presented to validate the proposed algorithm.

A novel algorithm for predicting target tracks through obscurations is introduced. This prediction method uses radar ground track indicators and the hidden transfer function (HTF) to predict future target locations. The HTF method is described in detail, and results provided that quantify track accuracy, forecast accuracy, and the percentage of tracks exiting an obscuration occurring that occur within the forecasted region. Five different classifier methods are shown for labeling short segments of track history. Each classifier method is scored and significance testing used to determine that the Data Model and SMART lookup table (LUT) were significantly better than the other classifier approaches.

Multi-target filtering for closely-spaced targets leads to degraded performance with respect to single-target filtering solutions, due to measurement provenance uncertainty. Soft data association approaches like the probabilistic data association filter (PDAF) suffer track coalescence. Conversely, hard data association approaches like multiplehypothesis tracking (MHT) suffer track repulsion. We introduce the stochastic data association filter (SDAF) that utilizes the PDAF weights in a stochastic, hard data association update step. We find that the SDAF outperforms the PDAF, though it does not match the performance of the MHT solution. We compare as well to the recentlyintroduced equivalence-class MHT (ECMHT) that successfully counters the track repulsion effect. Simulation results are based on the steady-state form of the Ornstein-Uhlenbeck process, allowing for lengthy stochastic realizations with closely-spaced targets.

The Histogram Probabilistic Multi-Hypothesis Tracker (H-PMHT) is a parametric track-before-detect algorithm that has been shown to give good performance at a relatively low computation cost. Recent research has extended the algorithm to allow it to estimate the signature of targets in the sensor image. This paper shows how this approach can be adapted to address the problem of group target tracking where the motion of several targets is correlated. The group structure is treated as the target signature, resulting in a two-tiered estimator for the group bulk-state and group element relative position.

This paper presents a separate spatio-temporal filter based small infrared target detection method to address the sea-based infrared search and track (IRST) problem in dense sun-glint environment. It is critical to detect small infrared targets such as sea-skimming missiles or asymmetric small ships for national defense. On the sea surface, sun-glint clutters degrade the detection performance. Furthermore, if we have to detect true targets using only three images with a low frame rate camera, then the problem is more difficult. We propose a novel three plot correlation filter and statistics based clutter reduction method to achieve robust small target detection rate in dense sun-glint environment. We validate the robust detection performance of the proposed method via real infrared test sequences including synthetic targets.

In this paper we present a new algorithm for approximating the target-measurement association probabilities of the Joint Probabilistic Data Association Filter (JPDAF). This algorithm is designed to robustify the JPDAF against track coalescence which can greatly degrade the performance of the JPDAF and other approximate algorithms. It is based on the works of Roecker and the JPDAF* of Blom and Bloem. We compare our new algorithm with the two it is based on, as well as the "cheap JPDAF" and the Set JPDAF, and show that it offers a significant improvement in computational complexity over the JPDAF*, and improvement in tracking error over the Roecker algorithm. We compare their performance with respect to the Mean Optimal Subpattern Assignment (MOSPA) statistic in scenarios involving several closely-spaced targets. A consistency comparison of the various algorithms considered is also presented.

This paper is Part VIc of a comprehensive survey of maneuvering target tracking without addressing the so-called measurement-origin uncertainty. It provides an in-depth coverage of various approximate density-based nonlinear filters in discrete time developed particularly for handling the uncertainties induced by potential target maneuvers as well as nonlinearities in the dynamical systems commonly encountered in target tracking. An emphasis is given to more recent results, especially those with good potential for tracking applications.