This PDF file contains the front matter associated with SPIE Proceedings Volume 9092 including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

One of the desired capabilities for wide-area persistent ISR systems is to reliably locate and subsequently track the movement of targets within the field of view. Current wide-area persistent ISR systems are characterized by large pixel overall counts and very large fields of view. This leads to a large ground sample distance with few pixels-on-target. Locating targets under these constraints is extremely difficult due to the fact that the targets present very little detailed structure. In this paper we will present the application of rich image feature descriptors combined with advanced statistical target detection methodologies to the airborne ISR problem. We will demonstrate that these algorithms can reliably locate targets in the scene without relying on the target's motion to form a detection. This is useful in ISR application where it is desirable to be able to continuously track a target through stops and maneuvers.

Small target detection and classification is problematic. For targets that operate as part of a cluster, classification can be performed based on the characteristics of the cluster’s operations, instead of trying to identify an individual clustermember directly. This paper presents an algorithm for object identification based on comparing networks of point-topoint distances between features identified by an image feature detection algorithm. It discusses the alterations required to make the algorithm suitable for performing cluster-formation based characterization of small targets from point or near-point source data. An analysis of the algorithm’s performance is presented and it efficacy for this application assessed.

M&M Aviation has been developing and conducting Hostile Fire Indication (HFI) tests using potassium line emission sensors for the Air Force Visible Missile Warning System (VMWS) to advance both algorithm and sensor technologies for UAV and other airborne systems for self protection and intelligence purposes. Work began in 2008 as an outgrowth of detecting and classifying false alarm sources for the VMWS using the same K-line spectral discrimination region but soon became a focus of research due to the high interest in both machine-gun fire and sniper geo-location via airborne systems. Several initial tests were accomplished in 2009 using small and medium caliber weapons including rifles. Based on these results, the Air Force Research Laboratory (AFRL) funded the Falcon Sentinel program in 2010 to provide for additional development of both the sensor concept, algorithm suite changes and verification of basic phenomenology including variance based on ammunition type for given weapons platform. Results from testing over the past 3 years have showed that the system would be able to detect and declare a sniper rifle at upwards of 3km, medium machine gun at 5km, and explosive events like hand-grenades at greater than 5km. This paper will outline the development of the sensor systems, algorithms used for detection and classification, and test results from VMWS prototypes as well as outline algorithms used for the VMWS. The Falcon Sentinel Program will be outlined and results shown. Finally, the paper will show the future work for ATD and transition efforts after the Falcon Sentinel program completed.

In this study, an intuitive way for tracking flocks of birds is proposed and compared to simple cluster-seeking algorithm for real radar observations. For group of targets such as flock of birds, there is no need to track each target individually. Instead a cluster can be used to represent closely spaced tracks of a possible group. Considering a group of targets as a single target for tracking provides significant performance improvement with almost no loss of information.

As the availability and use of imaging methodologies continues to increase, there is a fundamental need to jointly analyze data that is collected from multiple modalities. This analysis is further complicated when, the size or resolution of the images differ, implying that the observation lengths of each of modality can be highly varying. To address this expanding landscape, we introduce the multiset singular value decomposition (MSVD), which can perform a joint analysis on any number of modalities regardless of their individual observation lengths. Through simulations, the inter modal relationships across the different modalities which are revealed by the MSVD are shown. We apply the MSVD to forensic fingerprint analysis, showing that MSVD joint analysis successfully identifies relevant similarities for further analysis, significantly reducing the processing time required. This reduction, takes this technique from a laboratory method to a useful forensic tool with applications across the law enforcement and security regimes.

In this paper we present an approach for tracking with a high-bandwidth active radar in long range scenarios with 3-D measurements in r-u-v coordinates. The 3-D low-process-noise scenarios considered are much more difficult than the ones we have previously investigated where measurements were in 2-D (i.e., polar coordinates). We show that in these 3-D scenarios the extended Kalman filter and its variants are not desirable as they suffer from either major consistency problems or degraded range accuracy, and most flavors of particle filter suffer from a loss of diversity among particles after resampling. This leads to sample impoverishment and 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 is shown to avoid this diversity problem while producing consistent results. The regularization is accomplished using a modified version of the Epanechnikov kernel.

In this work we study the problem of detecting and tracking challenging targets that exhibit low signal-to-noise ratios (SNR). We have developed a particle filter-based track-before-detect (TBD) algorithm for tracking such dim targets. The approach incorporates the most recent state estimates to control the particle flow accounting for target dynamics. The flow control enables accumulation of signal information over time to compensate for target motion. The performance of this approach is evaluated using a sensitivity analysis based on varying target speed and SNR values. This analysis was conducted using high-fidelity sensor and target modeling in realistic scenarios. Our results show that the proposed TBD algorithm is capable of tracking targets in cluttered images with SNR values much less than one.

We describe four distinct ways to avoid normalization of the probability density for particle flow. We have roughly 20 algorithms to compute particle flow, and the three best algorithms avoid computing the normalization of the conditional probability density of the state. We explain why explicit normalization often spoils the flow. This phenomenon has been noticed by other researchers for completely different applications (e.g., weather prediction), but apparently the benefits of avoiding normalization are not well known.

We explain why one of our favorite particle flows (method #18) requires special care for numerical integration. The root cause is stiffness of the flow. This can be mitigated with a small increase in computational complexity (a factor of three). There are many ways to do this, but some are much better than others. In particular, one can use a smaller step size for numerical integration, or use principal coordinates, or use a stiff ODE solver or an adaptive ODE solver or other methods. We show that the best methods are contrary to the standard advice given in textbooks.

We discuss a procedure to estimate the state of thrusting/ballistic endoatmospheric projectiles for the purpose of impact point prediction (IPP). The short observation time and the estimation ambiguity between drag and thrust in the dynamic model motivate the development of a multiple interacting multiple model (MIMM) estimator with various drag coefficient initializations. In each IMM estimator used, as the mode-matched state estimators for its thrusting mode and ballistics mode are of unequal dimension, an unbiased mixing is required. We explore the MIMM estimator with unbiased mixing (UM) using extended Kalman filter (EKF), unscented Kalman filter (UKF) and particle filter (PF). For 30 real trajectories, the IPP based on the MIMM-UM estimation approach is carried out with various sets of tuning parameters selected. The MIMM-UM-EKF, MIMM-UM-UKF and MIMM-UM-PF are compared based on the resulting IPP performance, estimator consistency and computational complexity.

In this work, two multitarget trackers - the Cardinalized Probability Hypothesis Density (CPHD) filter and the Recursive Random Sample Consensus (R-RANSAC) algorithm - were applied to three scenarios of the Video Verification of IDentity (VIVID) dataset provided by DARPA. The dataset consists of real video data of multiple cars observed from an unmanned aerial vehicle (UAV) and includes challenging situations such as dense traffic and occlusions. The same detector output was given to each tracker and the same metrics of performance were computed in order to ensure fair comparison of the two tracking approaches. The results show the CPHD did better overall, which was to be expected given that it is the more mature approach.

In the space surveillance tracking domain, it is often necessary to assess not only the covariance consistency or covariance realism of an object's state estimate, but also the realism (proper characterization) of its full estimated probability density function. In other words, there is a need for “uncertainty realism." We propose a new metric (applicable to any tracking domain) that generalizes the covariance realism metric based on the Mahalanobis distance to one that tests uncertainty realism. We then review various goodness-of-fit and distribution matching tests that exploit the uncertainty realism metric and describe how these tests can be applied to assess uncertainty realism in off-line simulations with multiple Monte-Carlo trials or on-line with real data when truth is available.

In tracking, many types of sensor data can be obtained and utilized to distinguish a particular target. Commonly, kinematic information is used for tracking, but this can be combined with identification attributes and parametric information passively collected from the targets emitters. Along with the standard tracking process (predict, associate, score, update, and initiate) that operates in all kinematic trackers, parametric data can also be utilized to perform these steps and provide a means for feature fusion. Feature fusion, utilizing parametrics from multiple sources, yields a rich data set providing many degrees of freedom to separate and correlate data into appropriate tracks. Parametric radar data can take on many dynamics to include: stable, agile, jitter, and others. By utilizing a running sample mean and sample variance a good estimate of radar parametrics is achieved. However, when dynamics are involved, a severe lag can occur and a non-optimal estimate is achieved. This estimate can yield incorrect associations in feature space and cause track fragmentation or miscorrelation.
In this paper we investigate the accuracy of the interacting multiple model (IMM) filter at estimating the first and second moments of radar parametrics. The algorithm is assessed by Monte Carlo simulation and compared against a running sample mean/variance technique. We find that the IMM approach yields a better result due to its ability to quickly adapt to dynamical systems with the proper model and tuning.

We propose a unified testing framework for assessing uncertainty realism during non-linear uncertainty propagation under the perturbed two-body problem of celestial mechanics, with an accompanying suite of metrics and benchmark test cases on which to validate different methods. We subsequently apply the testing framework to different combinations of uncertainty propagation techniques and coordinate systems for representing the uncertainty. In particular, we recommend the use of a newly-derived system of orbital element coordinates that mitigate the non-linearities in uncertainty propagation and the recently-developed Gauss von Mises filter which, when used in tandem, provide uncertainty realism over much longer periods of time compared to Gaussian representations of uncertainty in Cartesian spaces, at roughly the same computational cost.

We present a new approach to estimating the probability of each association in a 2D assignment problem defined by likelihood ratios. Our method divides the set of feasible hypotheses into clusters, and converts a collection of hypotheses into a collection of clusters containing them, reducing the variance of the estimate. Simulations show that our method often generates substantially more accurate probability estimates in less time than traditional methods. Our method can obtain reasonably accurate probabilities of association based on only the input cost matrix and single best candidate solution, eliminating the need for a K-best solution method or MCMC sampling.

The Probabilistic Multi-Hypothesis Tracker (PMHT) was developed in the early 1990s by Roy Streit and Tod Luginbuhl. Since that time many advances and improvements have been made to this elegant algorithm that is linearly efficient in processing as the number of targets, sensors, and clutter increases. This paper documents the many advances to the PMHT by several different contributors over the past two decades. The history continues and looks as promising as ever for this algorithm as we present the latest advancement—the Maximum Likelihood, Histogram Probabilistic Multi-Hypothesis Tracker (ML-HPMHT)—and the exciting results of this potential game-changer in tracking unresolved, dim targets in highly cluttered environments. This new algorithm, which we are calling the Quanta Tracking algorithm, detects and tracks with high accuracy targets that are unresolved in pixels or range bins.

The calculation of marginal association probabilities is the major computational bottleneck in the Joint Probabilistic Data Association Filter (JPDAF). In this paper, we investigate approximations for the marginal associations that simplify the (computational complex) original association model in order to obtain efficient algorithms. In this context, we first discuss the Bakhtiar-Alavi algorithm and the Linear Multitarget Integrated Probabilistic Data Association (LMIPDA) algorithm. Second, we propose a fast novel approximation that exploits systematic combinations of the JPDAF measurement model with the Probabilistic Multi-Hypothesis Tracker (PMHT) measurement model. The discussed methods are evaluated by means of a tracking scenario with a high number of closely-spaced targets.

Recentwork developed a novelmethod for determining tracking thresholds for theMaximumLikelihood ProbabilisticMulti- Hypothesis Tracker (ML-PMHT). Under certain “ideal” conditions, probability density functions (PDFs) for the peak points in the ML-PMHT log-likelihood ratio (LLR) due to just clutter measurements could be calculated. Analysis of these clutter-induced peak PDFs allowed for the calculation of tracking thresholds, which previously had to be donewith time-consumingMonte Carlo simulations. However, this work was done for a very specific case: the amplitudes of both target and cluttermeasurements followed Rayleigh distributions. The Rayleigh distribution is a very light-tailed distribution, and it can be overly optimistic in predicting that high-SNR measurements are target-originated. This work examines the case where the clutter amplitudes do not follow a Rayleigh distribution at all, but instead follow a K-distribution, which more accurately describes active acoustic clutter. This will provide a framework for determining accurate tracking thresholds for the ML-PMHT algorithm.

Integration of space based sensors into a Ballistic Missile Defense System (BMDS) allows for detection and tracking of threats over a larger area than ground based sensors [1]. This paper examines the effect of sensor bias error on the tracking quality of a Space Tracking and Surveillance System (STSS) for the highly non-linear problem of tracking a ballistic missile. The STSS constellation consists of two or more satellites (on known trajectories) for tracking ballistic targets. Each satellite is equipped with an IR sensor that provides azimuth and elevation to the target. The tracking problem is made more difficult due to a constant or slowly varying bias error present in each sensor's line of sight measurements. It is important to correct for these bias errors so that the multiple sensor measurements and/or tracks can be referenced as accurately as possible to a common tracking coordinate system. The measurements provided by these sensors are assumed time-coincident (synchronous) and perfectly associated. The line of sight (LOS) measurements from the sensors can be fused into measurements which are the Cartesian target position, i.e., linear in the target state. We evaluate the Cramér-Rao Lower Bound (CRLB) on the covariance of the bias estimates, which serves as a quantification of the available information about the biases. Statistical tests on the results of simulations show that this method is statistically efficient, even for small sample sizes (as few as two sensors and six points on the (unknown) trajectory of a single target of opportunity). We also show that the RMS position error is significantly improved with bias estimation compared with the target position estimation using the original biased measurements.

We discuss the multistatic target tracking problem in which the target is illuminated by Doppler tolerant waveforms (DTW). DTW introduce a bias to the true time-of-arrival, as a result of which the corresponding range measurement accuracy is degraded when this bias is ignored. An important part in the design of a tracking filter using nonlinear measurements is the initialization: the initial state estimate should be unbiased and the associated covariance should be consistent with the errors. This paper presents a maximum likelihood based method as well as a simpler iterative bias cancellation method to initialize a tracking filter from two observation frames. The iterative method makes it possible to obtain unbiased initial position and velocity estimates by canceling out the Doppler bias. Simulation results show that both estimators are efficient.