This PDF file contains the front matter associated with SPIE Proceedings Volume 8857 including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

The detection of distant objects in an image is relevant to applications in defense, security, and robotics. Successfully detecting objects of interest has been a common problem with respect to intelligent computer vision and has been studied quite thoroughly. Automatic target recognition (ATR) systems have been formulated and employed in numerous ways to tackle the difficulty of unsupervised targeting [1]. We attempt to describe a new approach in the selection of a potential target area or region of interest (ROI) in an image. A region of interest from a particular image is simply a subset of an image whose size is dependent upon the user or is chosen by an adaptive process. Each ROI should contain an object of interest or information that is pertinent to the user or system. Thus, the goal of our framework is reducing the amount of ROIs without the unintended loss of an object of interest. We implement the Fast Discrete Curvelet Transform (FDCT) as the first stage in this system’s framework in order to have a collection of sufficient Curvelet coefficients. The Curvelet coefficients sparsify the input image and by this sparsification, we are able to locate and extract ROIs. Once the ROIs have been extracted, they are then passed to the second stage of our framework which is the classification stage. In the classification stage it is verified whether or not each ROI has an object of interest. After an image has been processed by this framework, the output should be the successful detection of meaningful objects or targets contained in the image. We shall first introduce our previous work and give a brief explanation of sparse representation. Once the concept of sparsity is understood, we will then commence with the summarization of the Curvelet transform. Having established these concepts, the new architecture of our system will be presented, followed by the experimentation and results.

An examination of the application of Space Time Adaptive Processing (STAP) techniques to real, multi-channel, medium grazing angle, radar sea clutter data is undertaken and the detection performance is quantified against simulated moving maritime surface targets. The application of sub-optimal STAP approaches to the maritime radar detection problem is shown to be complicated by non-stationarity of sea clutter and rapid variations of the sea clutter spectrum due to transient wave activity. Observed performance gains from maritime STAP are much more limited than those observed for Ground Moving Target Indication (GMTI) due to the inherent spectral width of sea clutter and the slow Doppler velocities of maritime targets. Three sub-optimal STAP processing architectures are examined and PRI-Staggered Post-Doppler is shown to provide consistently superior detection performance for the data set in question.

Infrared (IR) imagery sequences are commonly used for detecting moving targets in the presence of evolving cloud clutter or background noise. This research focuses on slow moving point targets that are less than one pixel in size, such as aircraft at long ranges from a sensor. The target detection performance is measured via the variance estimation ratio score (VERS), which essentially calculates the pixel scores of the sequences, where a high score indicates a target is suspected to traverse the pixel. VERS uses two parameters – long and short term windows, which were predetermined individually for each movie, depending on the target velocity and on the clouds intensity and amount, as opposed to clear sky (noise), in the background. In this work, we examine the correlation between the sequences' spatial and temporal features and these two windows. In addition, we modify VERS calculation, to enhance target detection and decrease cloud-edge scores and false detection. We conclude this work by evaluating VERS as a detection measure, using its original version and its modified version. The test sequences are both original real IR sequences as well as their relative compressed sequences using our designated temporal DCT quantization method.

Data containing the radar signature of amoving person on the groundwere collected at ranges of up to 30 kmfroma moving airborne platform using the DRDC Ottawa X-bandWideband Experimental Airborne Radar (XWEAR). The human target radar echo returns were found to possess a characteristic amplitude modulated (AM) and frequency modulated (FM) signature which could be usefully characterized in terms of conventional AM and FM modulation parameters. Human detection performance after space time adaptive processing is frequently limited by false alarms arising from incomplete cancellation of large radar cross-section discretes during the whitening step. However, the clutter discretes possess different modulation characteristics from the human targets discussed above. The ability of pattern classification techniques to use this parameter measurement space to distinguish between human targets and clutter discretes is explored and preliminary results presented.

In this paper we present an approach for tracking with a high-bandwidth active sensor in very long range scenarios. We show that in these scenarios the extended Kalman filter is not desirable as it suffers from major consistency problems; and most flavors of particle filter 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 is shown to avoid this diversity problem while producing consistent results. The regularization is accomplished using a modified version of the Epanechnikov kernel.

We derive a new algorithm for particle flow with non-zero diffusion corresponding to Bayes’ rule, and we report the results of Monte Carlo simulations which show that the new filter is an order of magnitude more accurate than the extended Kalman filter for a difficult nonlinear filter problem. Our new algorithm is simple and fast to compute, and it has an especially nice intuitive formula, which is the same as Newton’s method to solve the maximum likelihood estimation (MLE) problem (but for each particle rather than only the MLE), and it is also the same as the extended Kalman filter for the special case of Gaussian densities (but for each particle rather than just the point estimate). All of these particle flows apply to arbitrary multimodal densities with smooth nowhere vanishing non- Gaussian densities.

The problem of estimating the state of thrusting/ballistic endoatmospheric projectiles moving in 3-dimensional (3-D) space using 2-dimensional (2-D) measurements from a single passive sensor is investigated. The location of projectile’s launch point (LP) is unavailable and this could significantly affect the performance of the estimation and the IPP. The LP altitude is then an unknown target parameter. The estimability is analyzed based on the Fisher Information Matrix (FIM) of the target parameter vector, comprising the initial launch (azimuth and elevation) angles, drag coefficient, thrust and the LP altitude, which determine the trajectory according to a nonlinear motion equation. The full rank of the FIM ensures that one has an estimable target parameters. The corresponding Cram´er-Rao lower bound (CRLB) quantifies the estimation performance of the estimator that is statistically efficient and can be used for IPP. In view of the inherent nonlinearity of the problem, the maximum likelihood (ML) estimate of the target parameter vector is found by using a mixed (partially grid-based) search approach. For a selected grid in the drag-coefficient-thrust-altitude subspace, the proposed parallelizable approach is shown to have reliable estimation performance and further leads to the final IPP of high accuracy.

In order to carry out data fusion, registration error correction is crucial in multisensor systems. This requires estimation of the sensor measurement biases. 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. This paper provides a solution for bias estimation for the minimum number of passive sensors (two), when only targets of opportunity are available. The sensor measurements are assumed time-coincident (synchronous) and perfectly associated. Since these sensors provide only line of sight (LOS) measurements, the formation of a single composite Cartesian measurement obtained from fusing the LOS measurements from different sensors is needed to avoid the need for nonlinear filtering. We evaluate the Cramer-Rao Lower Bound (CRLB) on the covariance of the bias estimate, i.e., the 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 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.

Herein we describe theory and algorithms for detecting covariance structures in large, noisy data sets. Our work uses ideas from matrix completion and robust principal component analysis to detect the presence of low-rank covariance matrices, even when the data is noisy, distorted by large corruptions, and only partially observed. In fact, the ability to handle partial observations combined with ideas from randomized algorithms for matrix decomposition enables us to produce asymptotically fast algorithms. Herein we will provide numerical demonstrations of the methods and their convergence properties. While such methods have applicability to many problems, including mathematical finance, crime analysis, and other large-scale sensor fusion problems, our inspiration arises from applying these methods in the context of cyber network intrusion detection.

Detection of unusual trajectories of moving objects (e.g., people, automobiles, etc.) is an important problem in many civilian and military surveillance applications. In this work, we propose a multi-objective evolutionary algorithms and rough sets-based approach that breaks down 2-dimensional trajectories into a set of additive components, which then can be used to build a classifier capable of recognizing typical, but yet unseen trajectories, and identifying those that seem suspicious.

Pixel-size effects on the probability of detection and on measurement extraction accuracy (the position measurement noise variance) for point sources in the focal plane (FP) of an optical sensor are analyzed from a general target tracking perspective. The analysis uses the point spread function of the optical sensor, which causes distant targets | point sources | to appear as a spatially extended pattern in the FP. Measurement extraction, both via Maximum Likelihood Estimation (MLE) and pixel detection centroiding is examined. The impact of pixel size, target strength (SNR) and target motion uncertainty (“maneuvering index") on tracking accuracy are quantified theoretically and verified by simulations.

MLPDA (Maximum Likelihood Probabilistic Data Association) has drawn attention as an effective target track extraction algorithm in high false density environments. In this algorithm, the target track is estimated as the maximum likelihood state vector, by using multiple observation frames that include the target signal and many false signals. The track is confirmed whether it is the true target or not, by comparing its likelihood with a given track confirmation threshold. However, when the target signals are lost at several frames, the conventional MLPDA deteriorates the track estimation accuracy due to false signals in frames without the target signal. In this paper, we propose multiple hypothetical frame selection MLPDA, which can extract the target track under the situation where the target signals are lost in several frames. Specifically, a batch of stored frames is first selected for track extraction. If the track is not confirmed, our algorithm offers multiple frame selection hypotheses where some frames are assumed to be the frames without the target signal and the other frames include the target signal. The track is extracted under these hypotheses, respectively, and the most likely hypothesis is accepted. If all hypotheses are rejected, our proposed method generates hypotheses that increase the number of frames without the target signal, and verifies them again. Furthermore, the hypotheses that have likelihoods above a given threshold are retained in order to modify the wrong frame selection later. Simulation results show the validity of our proposed method.

We develop tractable solutions to the problem of controlling the directions of 2-D directional sensors for maximizing information gain corresponding to multiple targets in 2-D. The target locations are known with some uncertainty given by a joint prior distribution (Gaussian). A sensor generates a (noisy) measurement of a target only if the target lies within the field-of-view of the sensor, and the measurements from all the sensors are fused to form global estimates of target locations. This problem is hard to solve exactly - the computation time increases exponentially with the number of sensors. We develop heuristic methods to solve the problem approximately and provide lower and upper bounds on the optimal information gain. We improve the solutions from these heuristic approaches by formulating the problem as a dynamic programming problem and solving it using a rollout approach.

This paper presents a robust approach to the problem of localizing multiple emitters in cluttered environments using a network of low-cost sensors. A pseudo maximum likelihood (PML) function, whose peaks strongly indicate the number and locations of the true emitters, is generated from available sensor measurements. Localization is performed by first iteratively associating the individual sensor measurements to the PML peaks and then generating true maximum likelihood (ML) estimates of the emitter locations based on the associated measurements. These emitter location estimates are fed to an EKF-based, decentralized architecture for initiating, sustaining and sharing local track activity throughout the sensor network. The iterative PML approach is shown to be efficient with computational complexity that is linear in the number of measurements. Additionally, simulation results illustrate the ability of the proposed approach to generate accurate emitter location estimates in cluttered environments for generating and sustaining all track activity in a fully decentralized sensor network.

In this paper, we address the problem of passive tracking of multiple targets with the help of images obtained from passive infrared (IR) platforms. Conventional approaches to this problem, which involve thresholding, measurement detection, data association and filtering, encounter problems due to target energy being spread across multiple cells of the IR imagery. A histogram based probabilistic multi-hypothesis tracking (H-PMHT) approach provides an automatic means of modeling targets that are spread in multiple cells in the imaging sensor(s) by relaxing the need for hard decisions on measurement detection and data association. Further, we generalize the conventional HPMHT by adding an extra layer of EM iteration that yields the maximum likelihood (ML) estimate of the number of targets. With the help of simulated focal plane array (FPA) images, we demonstrate the applicability of MLHPMHT for enumerating and tracking multiple targets.

The goal of the present study is to develop a S/W that predicts infrared signals from objects synthetically by considering the internal and external factors, such as surface properties, internal heat sources, solar irradiations, atmospheric temperature, etc. As a part of developing a S/W, this paper contains some results of the measured and simulated IR signals for comparison and validation. The scaled model ship used in this study is made of 3mm-thick steel and 1.5m long. It has virtual internal heat sources that are made of a brass block with a heater. The experiments are performed under an indoor environment to exclude the external effects such as solar irradiance, wind effect, etc. The radiance from the ship is mainly due to the surface temperature because the effects of solar and wind are excluded. Also the IR signals are collected by the IR cameras which are operated at MWIR(3.7~5.1μm ) and LWIR(7.5~9.1 μm), respectively. At the same time, the atmospheric temperature and the relative humidity are measured to use for input conditions for calculation. The surface temperature and the IR signals within the two wavelengths are calculated by using the S/W developed in this study considering the 3D ship model, the internal heater temperature, and the atmospheric conditions. The temperature and IR signals from the scaled model ship obtained from measurements and from the S/W developed in this study are compared each other by showing fairly good agreements with each other.