A particular method of detecting unresolved targets using simulated monopulse radar data is examined in detail. The system is assumed to be correctly calibrated i.e. the decision boundary is calculated based on the true values governing the hypothesis that only a single target is present in the range cell. The system performance is analyzed under varying values for target ranges, angles between the beam pointing direction and the actual off-boresight angle of the targets, waveform power and number of pulses. It is shown that these parameters have a pronounced impact on the Boundary, Metric and Decision Surfaces. The False Alarm probability for a single target as a function of waveform power is considered, as also are the detection probabilities when two targets are present. The important issue of locating the decision point on the Boundary Surface is briefly discussed.

Infrared sensors and advanced signal processing are used to detect small (or point) targets in highly cluttered and noisy environments. In this paper, a wavelet detection algorithm and tracking of small targets in clutter will be discussed. A new registration algorithm based on optical flow estimates with matched subspace detectors against small maneuverable targets is also discussed. Both detectors incorporate adaptive constant false alarm rate (CFAR) detection statistics. Simulation of the detection and tracking algorithms using an unclassified database with a helicopter target and platform for the video cameras is summarized.

This paper presents a Fourier transform model of the aliasing effects caused by two-dimensional sensor undersampling. The SNR probability distribution at the sensor output is numerically generated and compared with simulation results. This distribution is needed for estimating the track score gain when SNR is used as a track feature. Modeling of aliasing makes it possible to calculate the sensor mean signal to noise ratio (SNR), the sensor radiometric measurement precision (RMP) and the sensor position measurement precision (PMP) without the use of Monte Carlo simulations. An example of a sensor design trade is presented in which the detector size is maximized with respect to ROC performance.

In this paper, we analyze features of radar returns from moving targets, introduce the basic concept of time-frequency-Radon transforms, describe the Radon transform for line feature detection, discuss their applications to detection of multiple moving targets in clutter, and demonstrate two examples of moving target detection using simulated radar data.

A new method of detection and determination of parameters for weakly reflecting and transparent moving objects is proposed. The method makes use of the coherent radiation scattered from inhomogeneous background surrounding these objects. It is based on pulsed coherent illumination of the objects and the background by two separated coherent sources, registering for each source the intensity of radiation scattered back by the background, and determining the difference between the time-averaged intensity squared and the square of the time-averaged intensity for various averaging times. The method proposed enables one to detect weakly reflecting and transparent objects placed at an arbitrary distance and to determine coordinates, sizes, and velocities of these objects. The probability of object detection is calculated versus variation and correlation radius of the background inhomogeneities under the assumption that the scattered field has Gaussian statistics. The paper presents a fast shifting algorithm of processing the signal received by means of the proposed method. The algorithm is based on illuminating the objects and the background by repeated pulses, storing the intensities of the reflected pulses and taking a time average of stored intensities squared and stored intensities in intervals between neighboring reflected pulses. Application of the proposed method for detecting clusters of weakly reflecting and transparent moving polluting particles in inhomogeneous environment and for determining position of clusters and concentration, average size, and average velocity of these particles is discussed.

In the presence of sea-surface multipath monopulse radar signals from a low elevation target have three alternative paths in addition to the direct (radar-to-target) path due to reflections from the sea surface. The specular reflection causes significant signal fading. The diffuse reflection causes an approximately constant bias to the in-phase component of the monopulse ratio, which is the standard extractor of the direction of arrival (DOA) in the monopulse processing. The diffuse reflection also causes higher standard deviation to the in-phase component of the monopulse ratio. In this paper we propose a maximum likelihood (ML) angle extraction technique for low elevation targets of known average signal strength having a Rayleigh fluctuation. The results show that this method reduces the error of the estimated angle compared to the conventional monopulse ratio estimator. Subsequently, the ML angle extractor is modified for the unknown average signal strength case. This modified angle extractor has only a small performance degradation compared with the known average signal strength case, but it performs much better than the monopulse ratio based estimator. This angle extractor reduces the root mean square error (RMSE) by more than 50% in the signal processing stage when used in a low flying target (sea skimmer) tracking scenario.

This paper develops a new algorithm for high range resolution (HRR) radar centroid processing for scenarios where there are closely spaced objects. For range distributed targets with multiple discrete scatterers, HRR radars will receive detections across multiple range bins. When the resolution is very high, and the target has significant extent, then it is likely that the detections will not occur in adjacent bins. For target tracking purposes, the multiple detections must be grouped and fused to create a single object report and a range centroid estimate is computed since the detections are range distributed. With discrete scatterer separated by multiple range bins, then when closely spaced objects are present there is uncertainty about which detections should be grouped together for fusion. This paper applies the EM algorithm to form a recursive measurement fusion algorithm that segments the data into object clusters while simultaneously forming a range centroid estimate with refined bearing and elevation estimates.

This paper shows that increasing SNR is not necessarily sufficient to improve ROC performance, that is, increase the probability of detection for a given false alarm probability. The general optimal detector is derived with no assumptions on the stationarity of the input noise process or the form of the input signal model.

In this paper the initial results are presented of a recursive filter based approach to Track Before Detect (TBD) for surveillance radar. Because of the relatively low update rate and large amount of data per scan, brute force methods such as the velocity filter banks or Hough transform for electro-optical sensors are not feasible. We therefore have designed a recursive algorithm that integrates the measured signal strength only for the most likely target trajectory. Furthermore, this algorithm is only started for those radar cells in a scan that have exceeded a preselection threshold. As will be shown, the use of a low preselection threshold significantly reduces the number of radar cells to be considered at a negligible performance reduction. The feasibility of the proposed algorithm is demonstrated through simulations. Further research planned and possible extensions to the initial approach are discussed.

The similarity between the multiple-target radar ranging problem and the multi-user detection problem in CDMA is drawn: in CDMA, users' bits modulate distinct but correlated signature signals; while, in radar, the bits are range-bin occupancies and the signatures correspond to the known transmitted signal translated to be centered on the appropriate range bin. The analogy is useful: there has been a great deal of recent experience in CDMA, and one of the best and fastest algorithms uses a variant of probabilistic data association (PDA, the target-tracking philosophy). PDA can be augmented by group decision feedback (GDF) -- another idea from CDMA -- to refine the target delay estimates; and finally minimum description length (MDL) is applied to estimate the number of targets. Simulation examples are given to illustrate the resolution of closely spaced targets within what would normally be thought the same range bin. Its performance is also compared with the Cramer-Rao lower bound (CRLB) and the alternating projection (AP) algorithm.

The problem of detecting whether or not a signal is present is often encountered. Several detection strategies lead to a likelihood ratio test against a threshold. In a simple setting explicit expressions for the likelihood can be obtained. However when the signal to be detected is generated by a nonlinear, non-Gaussian dynamical system it is in general impossible to obtain an expression for the probability of the signal under the hypothesis that it is present. Recently, so called particle filters have been proposed to solve nonlinear, non-Gaussian filtering problems. In this paper we show that the filtering solution obtained by a particle filter can be used to construct the likelihood ratio, needed to perform the likelihood ratio test for detection. Here we will show that different detection schemes can be used. These schemes have in common that they use the output of a particle filter for the purpose of detecting the possible presence of a target. Furthermore we will go into aspects that are of importance when actually building such a particle filter based detector.

Since the time and energy of phased array radars are under great demand in modern combat systems, methods that conserve those resources are very important. Two opportunities for conserving radar resources that have not been fully exploited when tracking closely-spaced objects with currently deployed systems are revisit time selection (i.e., time to make a measurement) and beam boresight placement. While these two functions are somewhat coupled, this paper addresses only the problem of beam pointing. Previously, a methodology for track management for phased array radars hinged on the concept of organizing tracks into, so called, dwell groups that included closely-spaced targets that could be illuminated with a single beam. Pointing angle for a dwell group was determined using a geometry-based approach. While the geometry-based approach was useful in improving the entire track management function, it was known to be sub-optimal in that the detection characteristics of the targets were not considered. This paper addresses an improved methodology for assigning membership in dwell groups and selecting dwell pointing angles.

This paper addresses scheduling algorithms to determine optimum utilization of an Airborne Early Warning (AEW) radar timeline resource based on radar constraints. The operation of an AEW surveillance radar in dense overland environments along with the presence of low altitude as well as highly maneuvering targets make detection and tracking a very complex one. A mechanically rotating antenna with electronic scanning capability addresses this problem. Not only does it provide maximum gain in the boresight direction of the antenna, but also the flexibility to focus energy and provide higher update rates at given sectors and selected targets. With the advent of electronic scanning, an efficient means of utilizing the radar timeline and waveforms with the available radar resources is required. To do this, a radar resource management concept is required for future AEW electronic scanning surveillance systems. This paper studies the resource management problem for an antenna with electronic scanning capabilities without rotation. The timeline is formulated in terms of radar dwells and revisit time constraints specified for each surveillance sector. A dwell is defined as radar time on target or angular position and revisit time is defined as the time between radar updates of a particular target or angular position. The methodology provides a criterion for determining if a feasible schedule exists that satisfies the dwell and revisit time constraints as well as methods for computing such schedules. The investigation includes the structure of optimal schedules and the complexity of the problem. Several solution techniques have been developed. The first algorithm developed is exact and it is based on dynamic programming. Since the problem is NP-hard, this algorithm is efficient for a small number of sectors. In order to solve medium and large size problems, heuristic approaches have been pursued. The heuristic developed is based on constrained semi-Markov decision processes. First, a relaxed version of the problem utilizing average re-visit time constraints is used rather than solving the problem in a rigorous way. Search methods are then used to find a rigorous solution.

A model is presented for predicting classification performance for systems having a large population of classes. The cases of large and small training set size for each class are treated separately. A method is proposed for measuring classification performance as the mean ranking statistic (rho) _E which is derived from the average information content h_E of the system feature vector, which is in turn derived from the system covariance matrices ((Sigma) _W, (Sigma) _B). This method for predicting (rho) _E is applied to the large training set case (case 1), explaining why performance is not compromised but improved by adding noisy features. The method is extended for predicting performance in the more difficult small training set case (case 2), explaining why performance may be compromised by the addition of noisy features in that situation.

Since the early 1990s, significant research has been done on a relatively new algorithm called the Probabilistic Multi-Hypothesis Tracker (PMHT). The majority of this research has concluded that there are a few weaknesses with this approach to tracking targets in the presence of clutter. First, the number of targets that are being tracked needs to be known a priori. Second, in order for the algorithm to operate properly, a very good initiation must be performed. Without a very close initiation, the PMHT usually fails to lock on to the target correctly. To address both of these issues, a hybrid approach is proposed. This hybrid approach will use a Multi-Hypothesis Tracking (MHT) algorithm to initiate new tracks and to continue tracking them until a track is stable. Then it will hand these tracks off to the PMHT to maintain. The MHT is very good at initiating new tracks, and the PMHT is best at maintaining multiple tracks because the algorithm's complexity with tracking additional targets grows linearly as opposed to exponentially.

A group target is a group of individual targets which are part of some larger military formation (battalion, brigade, tank column, aircraft, carrier group, etc.). Force aggregation (also known as Level 2 fusion or situation assessment) is the process of detecting, tracking, and identifying group targets. In an AeroSense Conference paper last year I proposed a theoretically unified and potentially practical approach to force aggregation based on finite-set statistics (FISST). Cluster processes were used to model the force aggregation problem. An optimal (but computationally intractable) Bayes filter was derived for force aggregation. Potential computational tractability was achieved by generalizing the concept of a first-order multitarget moment filter to group targets. Laser year, lack of space prevented me from explicitly describing the prediction and correction equations for this group-PHD filter. I do so in this paper, generalizing the PHD filter to include spontaneously generated targets. I also take this opportunity to respond to some recent published criticisms of FISST.

Recent work has been conducted to develop group tracking algorithms that identify and track multiple targets. One of the characteristics of the group tracking algorithms is the ability to correctly identify the target. If enough evidence has been accumulated to identify the target, the algorithms perform well. However, in the case of spurious measurements and obscured targets, the target identity may not be completely realizable. For the case in which the target identity is not discerned, it is important to classify the target based on some methodology to aid the user. Such a classification could be an allegiance so that when the algorithm groups targets, the information is useful to the human. One sensor that is ideal for the scenario is an Identify Friend Foe Neutral (IFFN) sensor which can classify the target allegiance. By incorporating an IFFN sensor in the GRoup IMM-JBPDAF Tracker (GRIT) algorithm, results show that when identity information is not available, target classification is realizable with allegiance features. Results are simulated for a high-range resolution radar (HRR) and an IFFN sensor and a 29% reduction in the computational classification due to the presence of clutter.

The telerobotic assembly of space-station components has become the method of choice for the International Space Station (ISS) because it offers a safe alternative to the more hazardous option of space walks. The disadvantage of telerobotic assembly is that it does not necessarily provide for direct arbitrary views of mating interfaces for the teleoperator. Unless cameras are present very close to the interface positions, such views must be generated graphically, based on calculated pose relationships derived from images. To assist in this photogrammetric pose estimation, circular targets, or spots, of high contrast have been affixed on each connecting module at carefully surveyed positions. The appearance of a subset of spots must form a constellation of specific relative positions in the incoming image stream in order for the docking to proceed. Spot positions are expressed in terms of their apparent centroids in an image. The precision of centroid estimation is required to be as fine as 1/20th pixel, in some cases. This paper presents an approach to spot centroid estimation using cross correlation between spot images and synthetic spot models of precise centration. Techniques for obtaining sub-pixel accuracy and for shadow and lighting irregularity compensation are discussed.

The standard approach to tracking a single target in clutter, using the Kalman filter or extended Kalman filter, is to gate the measurements using the predicted measurement covariance and then to update the predicted state using probabilistic data association. When tracking with a particle filter, an analog to the predicted measurement covariance is not directly available and could only be constructed as an approximation to the current particle cloud. A common alternative is to use a form of soft gating, based upon a Student's-t likelihood, that is motivated by the concept of score functions in classical statistical hypothesis testing. In this paper, we combine the score function and probabilistic data association approaches to develop a new method for tracking in clutter using a particle filter. This is done by deriving an expected likelihood from known measurement and clutter statistics. The performance of this new approach is assessed on a series of bearings-only tracking scenarios with uncertain sensor location and non-Gaussian clutter.

In this paper we consider a nonlinear bearing-only target tracking problem using three different methods and compare their performances. The study is motivated by a ground surveillance problem where a target is tracked from an airborne sensor at an approximately known altitude using depression angle observations. Two nonlinear suboptimal estimators, namely, the extended Kalman Filter (EKF) and the pseudomeasurement tracking filter are applied in a 2-D bearing-only tracking scenario. The EKF is based on the linearization of the nonlinearities in the dynamic and/or the measurement equations. The pseudomeasurement tracking filter manipulates the original nonlinear measurement algebraically to obtain the linear-like structures measurement. Finally, the particle filter, which is a Monte Carlo integration based optimal nonlinear filter and has been presented in the literature as a better alternative to linearization via EKF, is used on the same problem. The performances of these three different techniques in terms of accuracy and computational load are presented in this paper. The results demonstrate the limitations of these algorithms on this deceptively simple tracking problem.

For many dynamic estimation problems involving nonlinear and/or non-Gaussian models, particle filtering offers improved performance at the expense of computational effort. This paper describes a scheme for efficiently tracking multiple targets using particle filters. The tracking of the individual targets is made efficient through the use of Rao-Blackwellisation. The tracking of multiple targets is made practicable using Quasi-Monte Carlo integration. The efficiency of the approach is illustrated on synthetic data.

Boost phase missile tracking is formulated as a nonlinear parameter estimation problem, initialized with an unscented transformation, and updated with a scaled unscented Kalman filter.

Tracking large number of closely spaced objects is a challenging problem for any tracking system. In missile defense systems, countermeasures in the form of debris, chaff, spent fuel, and balloons can overwhelm tracking systems that track only individual objects. Thus, tracking these groups or clusters of objects followed by transitions to individual object tracking (if and when individual objects separate from the groups) is a necessary capability for a robust and real-time tracking system. The objectives of this paper are to describe the group tracking problem in the context of multiple frame target tracking and to formulate a general assignment problem for the multiple frame cluster/group tracking problem. The proposed approach forms multiple clustering hypotheses on each frame of data and base individual frame clustering decisions on the information from multiple frames of data in much the same way that MFA or MHT work for individual object tracking. The formulation of the assignment problem for resolved object tracking and candidate clustering methods for use in multiple frame cluster tracking are briefly reviewed. Then, three different formulations are presented for the combination of multiple clustering hypotheses on each frame of data and the multiple frame assignments of clusters between frames.

Measurements can arrive out-of-sequence at a central tracker due to varying data pre-processing times and communication delays in a multi-sensor target tracking system. A number of single-lag and multiple-lag out-of-sequence measurement (OOSM) filtering algorithms for the linear filtering problem are known in the research literature. In this paper, we present a multiple-lag nonlinear OOSM filtering algorithm based on an extension of the existing multiple-lag linear OOSM filtering algorithm Ground target tracking using multiple airborne ground moving target indicator (GMTI) radar sensors is an important problem in surveillance and precision tracking of ground moving targets. Sensor geometry with two nearly orthogonal GMTI sensors can significantly improve the position measurement accuracy with fast revisit times due to the narrow elliptical nature of the range and cross-range measurement error covariance matrix of a single sensor. We present numerical results for the multiple-lag nonlinear OOSM filtering algorithm using simulated GMTI measurements with nearly constant velocity motion in two dimensions. Our numerical results show that the results from the nonlinear OOSM algorithm are in close agreement with those obtained from the EKF using time-ordered GMTI measurements.

This paper presents the results of a study of tracking algorithms for maneuvering targets. The design focuses on alternative algorithms to track two-dimensional targets during maneuvers. The algorithms explored include a standard Kalman algorithm, an extended Kalman algorithm in which the target turn rate is an additional state variable, an interactive multiple model (IMM) algorithm consisting of two models with varying plant noise, a three-model IMM specifying three distinct target turn rates, and a constant gain alpha-beta filter. The IMM trackers tended to work the best in this study, with the three-model IMM performing best overall.

What underpins this vision as axiomatic is the mantra information is power. Besides the necessary requirement of information exchange networks with sufficient bandwidth and computational power to treat the data being passed around the network; algorithms are required to make sense of the data. It is estimation algorithms that turn the straw (data) into gold (information). Both proper execution and improvements in estimation algorithms are the enabling technology that facilitates the formation and usage of data across the envisioned warfare networks. We focus on some of the requirements that are driving the formation of these networks from a surface navy perspective in terms of estimation. We also discuss how these requirements focus the design of potentially new algorithms. We also discuss some of the crucial issues that may drive future requirements and algorithms.

Ballistic Missile Defense (BMD) effectiveness depends on a system's capability to acquire, track, identify, and engage threat missiles. The probability of a successful engagement can be improved by performing multiple-sensor data fusion, especially if the participating sensor systems are based on both radar frequency (RF) and infrared (IR) phenomenology. In this paper, we apply this observation to the Target Object Map (TOM) correlation problem for the standard configuration of a kill vehicle (with a single or multicolor IR seeker) receiving uplinks from a ground based radar. Specifically, we examine the application of a relative ranging technique that augments the angles-only track information of a passive IR sensor with non-parametric range-ranking of the threat complex. Since data association performance is significantly better for three-dimensional (3-D) matching that for two-dimensional (2-D) matching, the idea is to take advantage of relative range-ranking information of the threat complex to potentially improve performance. Numerous techniques that attempt to extract absolute range estimates from a passive IR sensor have been investigated by researchers in the BMD community and it is understood that range information allows for improved threat tracking, radiant intensity estimates, and data association performance. However, extracting absolute target range estimates from irradiance measurements is extremely difficult because of the presence of data uncertainties/ambiguities, environment and sensor noises, and small angular rates of tracked objects. Passive Relative Ranging (PRR) is distinct in that it focuses on the relative range-ranking of objects; knowledge that one object is closer than a second object, while not relevant for improving track or intensity estimation performance, can possibly improve the performance of sensor-to-sensor object assignment. The proposed PRR technique is based on the physical range-squared relationship between intensity and measured focal plane irradiance and the derived fact that threat objects at closer ranges with similar closing velocities have greater irradiance derivatives. This paper presents the theory behind PRR and presents preliminary performance results of the proposed PRR technique for statistically simulated data. Results are compared to those theoretically achievable, as determined by the Cramer-Rao Lower Bound (CRLB).

The integration of multiple sensors for the purpose of forming an integrated air picture has been intensely investigated in recent years. Assuming no sensor biases and minimal communication latencies, the optimal picture can be formed when all the sensor information is communicated to each network node. The state vectors for a given target at each node should be very similar. However, this does not occur in the presence of sensor bias which has an adverse effect on tracking performance. A method to account for the location, measurement, and attitude biases of the sensors must be employed to improve the accuracy of the target state estimates. This paper presents an absolute sensor alignment method to estimate the sensor bias in a multi-target environment. The output of the alignment process is used to compensate the sensor measurements employed in the tracking process. A comparison is made between the composite tracks generated using compensated and uncompensated measurements from multiple sensors.

Sensor data fusion has long been recognized as a means to improve target tracking. Common practice assumes that the sensors used are synchronous, i.e., they have identical data rate, measurements are taken at the same time, and have no communication delays between sensors platform and central processing center. Such assumptions are invalid in practice. Previous work of the authors dealt designing asynchronous track fusion filter that removes such assumptions when considering the multi-sensor target tracking case. This paper deals with the existence of a solution to the asynchronous track fusion problem for the case of three asynchronous sensors. In addition, the performance deterioration of the filter is analyzed as a function of the track fusion update rate for CV targets.

The methods used in the classification of multiple small targets can be very different from the methods commonly used in traditional pattern recognition. First, there may be characteristics of the features for each target class that can permit simpler computations than other features. In addition, in classifying targets, the target tracks are updated as new data becomes available and hence there can be a sequence of feature measurements that are available for the target classification process. In addition, with multiple targets, the a priori information may be in a form that make the classification processing for one target dependent on the classification processing of other targets. These aspects of target classification that make that processing different from traditional pattern recognition are the concern of this paper. To limit the length of the paper, the scope is restricted to classification tasks that allow the linear-Gaussian assumption to be used. Also, the data used in the classification process is restricted to features, i.e., no attributes, and the assumption is the tracker does not employ feature-aided tracking. While these assumptions simplify the discussion, the methods used could be modified to permit classification of a broad scope of classification tasks.

This paper describes the development of a general framework for the efficient management of multiple sensors in target tracking. The basis of the technique is to quantify, and subsequently control, the accuracy of target state estimation. The Posterior Cramer-Rao lower bound provides the means of achieving this aim by enabling us to determine a bound on the performance of all unbiased estimators of the unknown target state. The general approach is then to use optimization techniques to control the measurement process in order to achieve accurate target state estimation. We are concerned primarily with the deployment and utilization of a limited sensor resource. We also allow for measurement origin uncertainty, with sensor measurements either target generated or false alarms. We exploit previous work to determine a general expression for the Fisher Information Matrix in this case. We show that by making certain assumptions we can express the measurement uncertainty as a constant information reduction factor. This enables the Fisher Information Matrix to be calculated quickly, allowing Cramer-Rao bounds to be utilized for real-time, online sensor management.

Model-based, path estimation algorithms are commonly derived from a linear kinematic equation and an expansive observation model. The premise that the measurement region is all-inclusive simplifies the performance analysis, but it does not describe the relevant characteristics of some sensors. Most sensors have an active region. If the region contains the target, a noisy location measurement is made. Alternatively, no location measurement is returned. There is a clear tradeoff between measurement acuity and broad coverage. A sensor that adapts its active region to the most likely target location should achieve higher quality tracking and identification. However, effective utilization of this flexibility requires an accurate determination of the conditional distribution of the target-state. In this paper, the Gaussian Wavelet Estimator is used in an adaptive algorithm for sensor management. It is shown that the adaptive window provides performance that is superior to that achieved using a fixed window of comparable cover probability.

Particle filtering (PF) is a relatively new method to solve the nonlinear filtering problem, which is very general and easy to code. The main issue with PF is the large computational complexity. In particular, for typical low dimensional tracking problems, the PF requires 2 to 4 orders of magnitude more computer throughput than the EKF, to achieve the same accuracy. It has been asserted that the PF avoids the curse of dimensionality, but there is no formula or theorem that bounds or approximates the computational complexity of the PF as a function of dimension (d). In this paper, we will derive a simple back-of-the-envelope formula that explains why a carefully designed PF should mitigate the curse of dimensionality for typical tracking problems, but that it does not avoid the curse of dimensionality in general. This new theory is related to the fact that the volume of the d dimensional unit sphere is an amazingly small fraction of the d dimensional unit cube, for large d.

The problem of maintaining track on a primary target in the presence spurious objects is addressed. Recursive and batch filtering approaches are developed. For the recursive approach, a Bayesian track splitting filter is derived which spawns candidate tracks if there is a possibility of measurement misassociation. The filter evaluates the probability of each candidate track being associated with the primary target. The batch filter is a Markov-chain Monte Carlo (MCMC) algorithm which fits the observed data sequence to models of target dynamics and measurement-track association. Simulation results are presented.

Target tracking is usually performed using data from sensors such as radar, whilst the target identification task usually relies on information from sensors such as IFF, ESM or imagery. The differing nature of the data from these sensors has generally led to these two vital tasks being performed separately. However, it is clear that an experienced operator can observe behavior characteristics of targets and, in combination with knowledge and expectations of target type and likely activity, can more knowledgeably identify the target and robustly predict its track than any automatic process yet defined. Most trackers are designed to follow targets within a wide envelope of trajectories and are not designed to derive behavior characteristics or include them as part of their output. Thus, there is potential scope for both applying target type knowledge to improve the reliability of the tracking process, and to derive behavioral characteristics which may enhance knowledge about target identity and/or activity. In this paper we introduce a Bayesian framework for joint tracking and identification and give a robust and computationally efficient particle filter based algorithm for numerical implementation of the resulting recursions. Simulation results illustrating algorithm performance are presented.

Collectively moving ground targets are typical of a military ground situation and have to be treated as separate aggregated entities. For a long-range ground surveillance application with airborne GMTI radar we inparticular address the task of track maintenance for ground moving convoys consisting of a small number of individual vehicles. In the proposed approach the identity of the individual vehicles within the convoy is no longer stressed. Their kinematical state vectors are rather treated as internal degrees of freedom characterizing the convoy, which is considered as a collective unit. In this context, the Expectation Maximization technique (EM), originally developed for incomplete data problems in statistical inference and first applied to tracking applications by STREIT et al. seems to be a promising approach. We suggest to embed the EM algorithm into a more traditional Bayesian tracking framework for dealing with false or unwanted sensor returns. The proposed distinction between external and internal data association conflicts (i.e. those among the convoy vehicles) should also enable the application of sequential track extraction techniques introduced by Van Keuk for aircraft formations, providing estimates of the number of the individual convoy vehicles involved. Even with sophisticated signal processing methods (STAP: Space-Time Adaptive Processing), ground moving vehicles can well be masked by the sensor specific clutter notch (Doppler blinding). This physical phenomenon results in interfering fading effects, which can well last over a longer series of sensor updates and therefore will seriously affect the track quality unless properly handled. Moreover, for ground moving convoys the phenomenon of Doppler blindness often superposes the effects induced by the finite resolution capability of the sensor. In many practical cases a separate modeling of resolution phenomena for convoy targets can therefore be omitted, provided the GMTI detection model is used. As an illustration we consider the contribution of the proposed GMTI sensor model to the problem of early recognition of a stopping convoy.

Sharing data between two tracking systems frequently involves use of an object map: the transmitting system sends a frame of data with multiple observations, and the receiving system uses an assignment algorithm to correlate the information with its local observation data base. The usual prescription for this problem is an optimal assignment algorithm (such as JVC or auction) using a cost matrix based upon chi-squared distances between the local and remote observation data. The optimal assignment algorithm does not actually perform pattern matching, so this approach is not robust to large registration errors between the two systems when there exist differences in the number of observations held by both systems. Performance of a new assignment algorithm that uses a cost function including terms for both registration errors and track to track random errors is presented: the cost function explicitly includes a bias between the two observation sets and thus provides a maximum likelihood solution to the assignment problem. In practice, this assignment approach provides near perfect assignment accuracy in cases where the bias errors exceed the dimension of the transmitted object map and there exist mismatches in the numbers of observations made by the two systems. This performance extends to many cases where the optimal assignment algorithm methodology produces errors nearly 100% of the time. The paper includes the theoretical foundation of the assignment problem solved and comparison of achieved accuracy with existing optimal assignment approaches.

Preliminary tracking system design and analysis is typically done using simulated data in which the target truth is known and many techniques have been developed for evaluating performance under these conditions. However, there is a notable lack of any consistent approach for evaluating tracker performance for real data in which there may be an unknown number of targets of opportunity whose trajectories are not known. In addition, the background clutter/false alarm environment may be unknown so that an important analysis task is to determine the most accurate background models. This paper proposes a set of criteria for evaluating the tracks that are formed using real data collected in the field in the presence of an unknown number of targets of opportunity. These criteria include duration, update history, and measures of kinematic and data association consistency. A scoring method is developed and the use of these criteria for system design is discussed.

In this paper the track initiation problem is formulated as multiple composite hypothesis testing using maximum likelihood estimation with probabilistic data association (ML-PDA), an algorithm known to work under very low SNR conditions. This algorithm does not have to make a decision as to which measurement is target originated. The hypothesis testing is based on the minimum description length (MDL) criterion. We first review some well-known approaches for statistical model selection and the advantage of the MDL criterion. Then we present an approximate penalty in accounting for the model complexity to simplify the calculation of MDL. Finally, we apply the MDL approach for the detection and initiation of tracks of incoming tactical ballistic missiles in the exo-atmospheric phase using a surface based electronically scanned array (ESA) radar. The targets are characterized by low SNR, which leads to low detection probability and high false alarm rate. The target acquisition problem is formulated using a batch of radar scans to detect the presence of up to two targets. The ML-PDA estimator is used to initiate the tracks assuming the target trajectories follow a deterministic state propagation. The approximate MDL criterion is used to determine the number of valid tracks in a surveillance region. The detector and estimator are shown to be effective even at 4.4\,dB average SNR in a resolution cell (at the end of the signal processing chain).

We report an Optic fiber color sensor applied to measure the inside state of an airtight container, that will be able to indicate the life time of the container material. This measurement system consists of a cool light source, Y-branch optical fiber cable, photo-detectors and an information processing module which output the experimental result as the form of triple color primaries. Because the airtight container have only one hole, it only allows inserting the fiber, so applying the fiber optic sensor to detect is the better selection than other ways. The experiment results are discussed in detail. With simple structure, small volume, light weighty, the fiber optic color sensor can measure without effecting the inside vacuity of the airtight container. It is suitable to precisely remote control the measurement in the space of narrow and having complex atmosphere, and is in prospect of wide application.

The problem of detecting small target in IR imagery has attracted much research effort over the past few decades. As opposed to early detection algorithms which detect targets spatially in each image and then apply tracking algorithm, more recent approaches have used multiple frames to incorporate temporal as well as spatial information. They often referred to as track before detect algorithms. This approach has shown promising results particularly for detection of dim point-like targets. However, the computationally complexity has prohibited practical usage for such algorithms. This paper presents an adaptive, recursive and computation efficient detection method. This detection algorithm updates parameters and detects occurrence of targets as new frame arrived without storing previous frames, thus achieved recursiveness. Besides, the target temporal intensity change is modeled by two Gaussian distribution with different mean and variance. The derivation of this generalized model has taken account of the wide variation of target speed, therefore detects wider range of targets.

This paper develops methods for associating two sets of sensor tracks in the presence of missing tracks and translation bias. Key results include 1) extension of the maximum A Posteriori probability method of matching tracks to use feature information as well as kinematic information; 2) translation bias removal techniques that are computationally tractable for large numbers of tracks, and effective in the presence of missing tracks. These methods were evaluated by Monte Carlo simulation. The experimental results indicate that the maximum A Posteriori probability method with its adaptive threshold achieves close to its best performance for matching tracks without an additional threshold adjustment.

Track and tracklet fusion filtering is complicated because the estimation errors of tracks from two sources for the same target may be cross-correlated. This cross-correlation of these errors should be taken into account when designing the filter used to combine the track data. This paper addresses the various track and tracklet fusion methods and their impact on communications load and tracking characteristics. In track and tracklet fusion a sequence of measurements is processed at the sensor or platform level to form tracks; then sensor level track data (in the form of tracks or tracklets) for a target is distributed and fused with each other or with a global track. Track Fusion and Tracklet Fusion are also sometimes called Hierarchical Fusion, Federated Fusion, or Distributed Fusion. Track data can include features or other information useful for target classification. One characteristic of track and tracklet fusion that distinguishes one method from another is whether the local track data is combined or if global tracks are maintained and updated using the sensor track data. Another important issue is whether there is filter process noise to accommodate target maneuvers. Some filtering methods are designed for maneuvering targets and others are not. This paper enumerates the various track and tracklet fusion methods for processing data from distributed sensors and their impact on filter performance and communications load.

An algorithm fusion approach is presented to detect a short range target in the land environments with heavy clutters. There has been fundamental problems in extracting the small target from cluttered IR images in a land-based IR Sensor. One of the causes is that we do not have sufficient time to make a decision. In addition, high false alarm and low detection rate make it more difficult to track consecutively without disconnection. It is also found that the target occupies 5 to 7 pixels or more ones in sometimes when it approaches to the counter decision range. When the target is maneuvering the classical matched filter with Gaussian- shaped target assumption generates more false alarm and make it more difficult to track precisely. In order to overcome some of the difficulties, we combined two algorithms. One is the morphological nonlinear filter which is sensitive to the size of the target. The other is the classical matched filter applied to some of the clustered objects. One of the advantages of using morphological filter is that it requires less calculation time than the classical matched filter. It also provides more effective clues when the target is passing the decision range with predetermined target size. In order to reduce the false detection, two dimensional structuring elements are separated to one dimensional elements and applied opening minus closing operation to remove the longitudinal and lateral line objects abundant in the land background. In addition, real time calculation of matched filter based on the object clustering is proposed to implement in real system.

This is the fourth part of a series of papers that provide a comprehensive survey of techniques for tracking maneuvering targets without addressing the so-called measurement-origin uncertainty. Part I and Part II deal with target motion models. Part III covers the measurement models and the associated techniques. This part surveys tracking techniques that are based on decisions regarding target maneuver. Three classes of techniques are identified and described: equivalent noise, input detection and estimation, and switching model. Maneuver detection methods are also included.

Image registration refers to the problem of spatially aligning two or more images. A challenging problem in this area is the registration of images obtained by different types of sensors. In general such images have different gray level characteristics and commonly used techniques such as those based on area correlations cannot be applied directly. On the other hand, contours representing the region boundaries are preserved in most cases. Therefore, contour based registration techniques are applicable to multimodal sensors. In this paper, various registration techniques based on subband decomposition and projection along x and y directions are introduced. The effect of binarization is investigated. Unknown translation and scaling parameters are computed using cross-correlation methods over the projections. Performance of the algorithms is compared.

Target selection is the task of assigning a value or priority to various targets in a scenario. This priority is usually determined by the threat the target poses on the defender in addition to its vulnerability to possible measures to be taken by the defender. In this study, we describe a target selection technique based on neural networks. The utility or value of each target is assumed to be an unknown function acting on certain features of the target such as size, intensity, speed and direction of movement. Neural networks used in the context of function estimation is a viable candidate for determining this unknown function for generating target priorities. Various neural network configurations are examined and simulation results are presented.