This PDF file contains the front matter associated with SPIE Proceedings Volume 9476 including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

Target classification algorithms have generally kept pace with developments in the academic and commercial sectors since the 1970s. However, most recently, investment into object classification by internet companies and various Human Brain Projects have far outpaced that of the defense sector. Implications are noteworthy. There are some unique characteristics of the military classification problem. Target classification is not solely an algorithm design problem, but is part of a larger system design task. The design flows down from a concept of operations (ConOps) and key performance parameters (KPPs). Inputs are image and/or signal data and time-synchronized metadata. The operation is real-time. The implementation minimizes size, weight and power (SWaP). The output must be conveyed to a time-strapped operator who understands the rules of engagement. It is assumed that the adversary is actively trying to defeat recognition. The target list is often mission dependent, not necessarily a closed set, and may change on a daily basis. It is highly desirable to obtain sufficiently comprehensive training and testing data sets, but costs of doing so are very high and data on certain target types are scarce. The training data may not be representative of battlefield conditions suggesting the avoidance of highly tuned designs. A number of traditional and emerging target classification strategies are reviewed in the context of the military target problem.

In this paper we investigate the problem of fusing a set of features for a discriminative visual tracking algorithm, where good features are those that best discriminate an object from the local background. Using a principled Mutual Information approach, we introduce a novel online feature selection algorithm that preserves discriminative features while reducing redundant information. Applying this algorithm to a discriminative visual tracking system, we experimentally demonstrate improved tracking performance on standard data sets.

Evaluation of signature properties of military equipment is very important. It is crucial to apply the proper method out of many possible approaches, based on amongst others ranking by probability of detection, detection time, and distance to target, which have been carried out by various countries. In this paper we present results from camouflage pattern assessments utilising two different approaches, based on human observers (detection time) and simulations (CAMAELEON). CAMAELEON ranks camouflaged targets by their local contrast, orientation and spatial frequency, mimicking the human eye’s response, and is a rapid and low cost method for signature assessment. In our camouflage tests, human observers were asked to search for targets (in a natural setting) presented on a high resolution pc screen, and the corresponding detection times were recorded. In our study we find a good correspondence between the camouflage properties of the targets in most of our unique tests (scenes), but in some particular cases there is an interesting deviation. Two similar camouflage patterns (both were random samples of the pattern) were tested, and it seemed that the results depended on the way the pattern is attached to the test subject. More precisely, it may seem that high-contrast coloured patches of the pattern in the target outline were significantly different detected by humans compared to CAMAELEON. In this paper we discuss this deviation in the two signature evaluation methods and look at potential risks.

In this paper, we are proposing a method for the principled evaluation of scene understanding systems in a query-based framework. We can think of a query-based scene understanding system as a generalization of typical sensor exploitation systems where instead of performing a narrowly defined task (e.g., detect, track, classify, etc.), the system can perform general user-defined tasks specified in a query language. Examples of this type of system have been developed as part of DARPA’s Mathematics of Sensing, Exploitation, and Execution (MSEE) program. There is a body of literature on the evaluation of typical sensor exploitation systems, but the open-ended nature of the query interface introduces new aspects to the evaluation problem that have not been widely considered before. In this paper, we state the evaluation problem and propose an approach to efficiently learn about the quality of the system under test. We consider the objective of the evaluation to be to build a performance model of the system under test, and we rely on the principles of Bayesian experiment design to help construct and select optimal queries for learning about the parameters of that model.

Instantaneous frequency is an important characteristic of time-varying or nonstationary signals. The definition and interpretation of instantaneous frequency have been the subject of discussion and debate for decades. The most common approach is due to Gabor, whereby a specific complex signal, called the analytic signal, is associated with a given real signal by inverting the spectrum of the real signal over only the positive frequency axis; the instantaneous frequency is then taken to be the derivative of the phase. Other approaches for associating a particular complex signal to a given real signal, and hence obtaining different instantaneous frequencies, have also been proposed. One way to define the associated complex signal / instantaneous frequency is by imposing physical constraints, which we discuss. We also discuss the common interpretation of instantaneous frequency as the average frequency at each time, and point out when this interpretation holds, which is not usually the case. This leads to the question of what is the “average frequency at each time?” The answer, coupled with physical constraints on the complex signal representation, leads to a quadrature-AM / FM signal model. Finally, we consider methods that manipulate the poles and zeros of the signal to obtain a complex representation.

Target detection is the act of isolating objects of interest from the surrounding clutter, generally using some form of test to include objects in the found class. However, the method of determining the threshold is overlooked relying on manual determination either through empirical observation or guesswork. The question remains: how does an analyst identify the detection threshold that will produce the optimum results? This work proposes the concept of a target detection sweet spot where the missed detection probability curve crosses the false detection curve; this represents the point at which missed detects are traded for false detects in order to effect positive or negative changes in the detection probability. ROC curves are used to characterize detection probabilities and false alarm rates based on empirically derived data. It identifies the relationship between the empirically derived results and the first moment statistic of the histogram of the pixel target value data and then proposes a new method of applying the histogram results in an automated fashion to predict the target detection sweet spot at which to begin automated target detection.

This work examines joint anomaly detection and dictionary learning approaches for identifying anomalies in persistent surveillance applications that require data compression. We have developed a sparsity-driven anomaly detector that can be used for learning dictionaries to address these challenges. In our approach, each training datum is modeled as a sparse linear combination of dictionary atoms in the presence of noise. The noise term is modeled as additive Gaussian noise and a deterministic term models the anomalies. However, no model for the statistical distribution of the anomalies is made. An estimator is postulated for a dictionary that exploits the fact that since anomalies by definition are rare, only a few anomalies will be present when considering the entire dataset. From this vantage point, we endow the deterministic noise term (anomaly-related) with a group-sparsity property. A robust dictionary learning problem is postulated where a group-lasso penalty is used to encourage most anomaly-related noise components to be zero. The proposed estimator achieves robustness by both identifying the anomalies and removing their effect from the dictionary estimate. Our approach is applied to the problem of ship detection and tracking from full-motion video with promising results.

Vessel type classification in maritime imagery is a challenging problem and has applications to many military and surveillance applications. The ability to classify a vessel correctly varies significantly depending on its appearance which in turn is affected by external factors such as lighting or weather conditions, viewing geometry and sea state. The difficulty in classifying vessels also varies among different ship types as some types of vessels show more within-class variation than others. In our previous work, we showed that the bag of visual words" (V-BoW) was an effective feature representation for this classification task in the maritime domain. The V-BoW feature representation is analogous to the bag of words" (BoW) representation used in information retrieval (IR) application in text or natural language processing (NLP) domain. It has been shown in the textual IR applications that the performance of the BoW feature representation can be improved significantly by applying appropriate term-weighting such as log term frequency, inverse document frequency etc. Given the close correspondence between textual BoW (T-BoW) and V-BoW feature representations, we propose to apply several well-known term weighting schemes from the text IR domain on V-BoW feature representation to increase its ability to discriminate between ship types.

The paper considers a fundamental particle separation problem in microfluidic devices,e.g.,microchannels. It is expected that particles with different electric characteristics flow into the different microchannels to achieve the separation purpose. The movement of the particles inside the microchannels is recorded in video as data. The objective of the research is to obtain the trajectories of the particles, and eventually establish the relationship between the particles dynamic characteristics and their electric characteristics. This paper proposes a framework that consists of raw image segmentation and multiple target trackers(multiple hypothesis tracker or Gaussian mixture probability hypothesis density tracker) to obtain the tracks of the particles.

Morphological filtering, erosion and dilation on images, has been successfully applied to many image processing and pattern recognition fields. Landing site selection on the other hand, finding the best (safest) point to land with given obstacle distributions and aircraft shapes, has significant value for landing aircraft in tight environments. In this paper, we derive shape distance transform theory, and using this theory, we build a connection between morphological filtering and landing site selection which are traditionally treated as two different topics. From shape distance transform theory, we show that morphological filtering and land site selection can be implemented by shape distance transform. Thus, fast shape distance transform will provide fast implementation of morphological filtering and landing site selection. For convex polygon shape templates, we propose propagation techniques to compute shape distance transform. Then we introduce a new approach for faster morphological filtering and landing site selection. This new approach is independent of template sizes and its computational complexity is 0(1) which is much lower than 0(N) of the direct implementation, where N is the number of pixels in templates. For large templates, N is large, and the new approach is significantly faster than the direct implementation

This paper describes an approach to identify individuals with suspicious objects in a crowd. To accomplish this goal we define criteria for a suspicious individual we are searching for. The query image is declared to contain a suspicious individual if it satisfies these criteria. In our implementation we apply a well-known algorithm suite used in image retrieval, mobile visual search problems where the reference data base of images is stored in a hierarchical tree data structure. In many cases, the construction of such a hierarchical tree uses k-means clustering followed by geometric verification. However, the number of clusters is not known in advance, and sometimes it is randomly generated. This may lead to congested clustering which can cause problems in grouping large real-time data. To overcome this problem, in this work, we estimate the number of clusters using the Indian Buffet stochastic process. We present examples illustrating our method.

We explore the use of hyperdimensional manifolds on Automatic Target Recognition (ATR) using Synthetic Aperture Radar (SAR) imagery. Data that occupies within a hyperdimensional space can be exploited using measures constrained along the inherent structure (manifold). When compared to multiple manifolds representing different classes. associations can be made by utilizing these constrained measures and data clusters to assign the closest class. We also explore the use of sparsely estimated manifolds (limited training data) and its impact on ATR results on SAR imagery.

In active sensing such as in sonar and radar, target recognition is adversely impacted by target-like returns from non-target objects (i.e. clutter). Because the target and clutter returns are in general nonstationary, the application of linear time-varying (LTV) pre-filters has been suggested to enhance target classification. We apply a minimum probability of error (MPE) classifier with and without LTV filters to distinguish targets from clutter in active sonar data. Classification performance was improved with LTV filtering.

Most often, background subtraction and image segmentation methods use images or video captured using a single camera. However, segmentation can be improved using stereo images by reducing errors caused due to illumination fluctuations and object occlusion. This work proposes a background subtraction and image segmentation method for images obtained using a two camera stereo system. Stereo imaging is often employed in order to obtain depth information. On the other hand, the objective of this work is mainly to extract accurate boundaries of objects from stereo images, which are otherwise difficult to obtain. Improving the outline detection accuracy is vital for object recognition applications. An application of the proposed technique is presented for the detection and tracking of fish in underwater image sequences. Outline fish detection is a challenging task since fish are not rigid objects. Moreover, color is not necessarily a reliable means to segment underwater images, therefore, grayscale images are used. Due to these two reasons, and due to the fact that underwater images captured in non-controlled environments are often blurry and poorly illuminated, commonly used local correlation methods are not sufficient for stereo image matching. The proposed algorithm improves segmentation in several scenarios including cases where fish are occluded by other fish regions. Although the work concentrates on segmenting fish images, it can be employed in other underwater image segmentation applications where visible-light cameras are used.

We consider the issue of local values for deterministic and random signals. By local values we mean physical quantities that are functions of time such as instantaneous frequency. We discuss how to define instantaneous bandwidth and show that it always consists of two parts which have different physical meanings. One is due to the instantaneous amplitude modulation and the other to instantaneous frequency modulation. The issue of the proper definitions for the random case is considered and a number of examples are given.

This paper summarizes part of a study to address the issue of underwater automatic object detection and classification of mine-like objects by means of a sonar sensor. The ultimate goals were to develop methods to adaptively selects the optimum algorithms and their parameters as sensor parameters and environmental conditions change. For adaptation, the method exploits predictive performance models of target detection and classification in terms of sea state, sensor and environmental parameters, target detection and classification algorithms and their internal parameters. This paper is the first in a number of upcoming reports and describes a number of key exploitation algorithms that were used and their sample performance results. In the future, separate papers will address the performance estimation and adaptation aspects of this study.

Peaky template matching (PTM) is a special case of a general algorithm known as multinomial pattern matching originally developed for automatic target recognition of synthetic aperture radar data. The algorithm is a model- based approach that first quantizes pixel values into N_q = 2 discrete values yielding generative Beta-Bernoulli models as class-conditional templates. Here, we consider the case of classification of target chips in AWGN and develop approximations to image-to-template classification performance as a function of the noise power. We focus specifically on the case of a uniform quantization" scheme, where a fixed number of the largest pixels are quantized high as opposed to using a fixed threshold. This quantization method reduces sensitivity to the scaling of pixel intensities and quantization in general reduces sensitivity to various nuisance parameters difficult to account for a priori. Our performance expressions are verified using forward-looking infrared imagery from the Army Research Laboratory Comanche dataset.

Top-Hat transform is well known background suppression method used in small target detection. In this paper, we investigate various different Top-Hat transformation based small target detection approaches. All of the methods are implemented with their best parameter settings and applied to the same test image. The comparison among them is done in terms of three issues: 1. the detection performance (precision and false alarm rate), 2. the time requirement of the method and its usability for real time applications, 3. the number of parameters, which need user interaction. Results show that all of the algorithms require a prior knowledge of target size, which is either used as the structuring element size or as the threshold for post-processing. Algorithms, which use automatic approaches to select its parameters, are not generic to be applied to various images. But algorithms, which use adaptive methods for deciding on the threshold value, perform better than the others.

Typical supervised classification algorithms label inputs according to what was learned in a training phase. Thus, test inputs that were not seen in training are always given incorrect labels. Open set recognition algorithms address this issue by accounting for inputs that are not present in training and providing the classifier with an option to reject" unknown samples. A number of such techniques have been developed in the literature, many of which are based on support vector machines (SVMs). One approach, the 1-vs-set machine, constructs a slab" in feature space using the SVM hyperplane. Inputs falling on one side of the slab or within the slab belong to a training class, while inputs falling on the far side of the slab are rejected. We note that rejection of unknown inputs can be achieved by thresholding class posterior probabilities. Another recently developed approach, the Probabilistic Open Set SVM (POS-SVM), empirically determines good probability thresholds. We apply the 1-vs-set machine, POS-SVM, and closed set SVMs to FLIR images taken from the Comanche SIG dataset. Vehicles in the dataset are divided into three general classes: wheeled, armored personnel carrier (APC), and tank. For each class, a coarse pose estimate (front, rear, left, right) is taken. In a closed set sense, we analyze these algorithms for prediction of vehicle class and pose. To test open set performance, one or more vehicle classes are held out from training. By considering closed and open set performance separately, we may closely analyze both inter-class discrimination and threshold effectiveness.

Failure-free operation of solar panels is of fundamental importance for modern commercial solar power plants. To achieve higher power generation efficiency and longer panel life, a simple and reliable panel evaluation method is required. By using thermal infrared imaging, anomalies can be detected without having to incorporate expensive electrical detection circuitry. In this paper, we propose a solar panel defect detection system, which automates the inspection process and mitigates the need for manual panel inspection in a large solar farm. Infrared video sequences of each array of solar panels are first collected by an infrared camera mounted to a moving cart, which is driven from array to array in a solar farm. The image processing algorithm segments the solar panels from the background in real time, with only the height of the array (specified as the number of rows of panels in the array) being given as prior information to aid in the segmentation process. In order to “count” the number the panels within any given array, frame-to frame panel association is established using optical flow. Local anomalies in a single panel such as hotspots and cracks will be immediately detected and labeled as soon as the panel is recognized in the field of view. After the data from an entire array is collected, hot panels are detected using DBSCAN clustering. On real-world test data containing over 12,000 solar panels, over 98% of all panels are recognized and correctly counted, with 92% of all types of defects being identified by the system.

We present methods for accurately estimating and tracking instantaneous frequency and relative time delay of narrowband signal components. These methods are applied to the problem of estimating the location of an emitter from the signal(s) received by one or more receivers. Both instantaneous frequency estimation and time delay estimation are based on previously reported cross-spectral methods that have been applied successfully to a variety of signal processing problems. Accurate geolocation is accomplished by matching the Doppler characteristics of the received signal to Doppler characteristics estimated from the known emitter motion and possible emitter locations.

In two-dimensional electron systems with mobility on the order of 1,000 – 10,000 cm²/Vs, the electron scattering time is about 1 ps. For the THz window of 0.3 – 3 THz, the THz photon energy is in the neighborhood of 1 meV, substantially smaller than the optical phonon energy of solids where these 2D electron systems resides. These properties make the 2D electron systems interesting as a platform to realize THz devices. In this paper, I will review 3 approaches investigated in the past few years in my group toward THz devices. The first approach is the conventional high electron mobility transistor based on GaN toward THz amplifiers. The second approach is to employ the tunable intraband absorption in 2D electron systems to realize THz modulators, where I will use graphene as a model material system. The third approach is to exploit plasma wave in these 2D electron systems that can be coupled with a negative differential conductance element for THz amplifiers/sources/detectors.

This paper considers a problem of distributed function estimation in the case when sensor locations are modeled as Gaussian random variables. We consider a scenario where sensors are deployed in clusters with cluster centers known a priori (or estimated by a high performance GPS) and the average quadratic spread of sensors around the cluster center also known. Distributed sensors make noisy observations about an unknown parametric field generated by a physical object of interest (for example, magnetic field generated by a ferrous object and sensed by a network of magnetometers). Each sensor then performs local signal processing of its noisy observation and sends it to a central processor (called fusion center) in the wireless sensor network over parallel channels corrupted by fading and additive noise. The central processor combines the set of received signals to form an estimate of the unknown parametric field. In our numerical analysis, we involve a field shaped as a Gaussian bell. We experiment with the size of sensor clusters and with their number. A mean square error between the estimated parameters of the field and the true parameters used in simulations is involved as a performance measure. It can be shown that a relatively good estimate of the field can be obtained with only a small number of clusters. As the number of clusters increases, the estimation performance steadily improves. The results also indicate that, on the average, the number of clusters has more impact on the performance than the number of sensors per cluster, given the same size of the total network.

In this paper we develop two statistical rules for the purpose of detecting pulsars and transients using signals from phased array feeds installed on a radio telescope in place of a traditional horn receiver. We assume a known response of the antenna arrays and known coupling among array elements. We briefly summarize a set of pre-processing steps applied to raw array data prior to signal detection and then derive two detection statistics assuming two models for the unknown radio source astronomical signal: (1) the signal is deterministic and (2) the signal is a random process. The performance of both detectors is analyzed using both real and simulated data.

Matching facial images across electromagnetic spectrum presents a challenging problem in the field of biometrics and identity management. An example of this problem includes cross spectral matching of active infrared (IR) face images or thermal IR face images against a dataset of visible light images. This paper describes a new operator named Composite Multi-Lobe Descriptor (CMLD) for facial feature extraction in cross spectral matching of near-infrared (NIR) or short-wave infrared (SWIR) against visible light images. The new operator is inspired by the design of ordinal measures. The operator combines Gaussian-based multi-lobe kernel functions, Local Binary Pattern (LBP), generalized LBP (GLBP) and Weber Local Descriptor (WLD) and modifies them into multi-lobe functions with smoothed neighborhoods. The new operator encodes both the magnitude and phase responses of Gabor filters. The combining of LBP and WLD utilizes both the orientation and intensity information of edges. Introduction of multi-lobe functions with smoothed neighborhoods further makes the proposed operator robust against noise and poor image quality. Output templates are transformed into histograms and then compared by means of a symmetric Kullback-Leibler metric resulting in a matching score. The performance of the multi-lobe descriptor is compared with that of other operators such as LBP, Histogram of Oriented Gradients (HOG), ordinal measures, and their combinations. The experimental results show that in many cases the proposed method, CMLD, outperforms the other operators and their combinations. In addition to different infrared spectra, various standoff distances from close-up (1.5 m) to intermediate (50 m) and long (106 m) are also investigated in this paper. Performance of CMLD is evaluated for of each of the three cases of distances.

In the present paper we study the problem of weapon identification and threat assessment from a single image with a partially occluded weapon. This problem poses very severe restrictions. To successfully identify a weapon from its parts we extend the first firearm ontology with the meronymic (partonomic) principle which lets us distinguish parts of a gun (e.g., lock, barrel, stock, scope). Adding classes of meronymic information provides meta-data (necessary for threat assessment) and allows for fast and accurate search. Searching for a weapon is treated conceptually as searching for the sum of its parts. An expanding active contour and morphological techniques are applied to partition weapons and extract boundaries, and a minimal inscribed complex polygon. Finite numerical sequences are generated, from the extracted geometric features, and are used to label partonomic nodes and perform quick and accurate search. The paper reports experimental results on weapons partitioning and search.

We investigate the feasibility of negative photoresist as a structural material in metal-organic hybrid THz imaging detectors using SU-8. We will discuss design of metamaterials for MEMS-based terahertz (THz) thermal sensors and design and microfabrication process for building SU8-based MEMS THz focal plane arrays. Metamaterials of this kind, exhibiting absorption properties comparable to those of resonant metamaterials made using traditional thin films, coupled with the applicability of SU-8 as a structural material, offer possibilities for quick, simple microfabrication of focal plane arrays of THz imaging detectors. SU-8 is a low-cost material that can quickly be spun onto a substrate at a wide range of thicknesses and photolithographically patterned into a variety of structures. This removes the need for both PECVD deposition and plasma etching, dramatically increasing the speed and lowering the cost of production of such FPAs. We further investigate feasibility of use of such detectors as band translators rather than traditional bimaterial devices. Translators would be optically probed with an infrared (IR) camera. Individual pixels would absorb THz radiation, heat up and the thermal image would be projected onto an infrared camera, effectively translating the image from THz into IR.

The present approach combines data fusion from several sensor types to enhance the overall detection and classification performance. The fusion of different sensors is implemented at data and feature levels that results in enhanced target identification by the means of spatial spectral analysis.

This paper reviews recent advances in the double-graphene-layer (DGL) active plasmonic heterostructures for the terahertz (THz) device applications. The DGL consists of a core shell in which a thin tunnel barrier layer is sandwiched by the two GLs being independently connected with the side contacts and outer gate stack layers at both sides. The DGL core shell works as a nano-capacitor, exhibiting inter-GL resonant tunneling (RT) when the band offset between the two GLs is aligned. The RT produces a strong nonlinearity with a negative differential conductance in the DGL current-voltage characteristics. The excitation of the graphene plasmons by the THz radiation resonantly modulates the tunneling currentvoltage characteristics. When the band offset is aligned to the THz photon energy, the DGL structure can mediate photonassisted RT, resulting in resonant emission or detection of the THz radiation. The cooperative double-resonant excitation with structure-sensitive graphene plasmons gives rise to various functionalities such as rectification (detection), photomixing, higher harmonic generation, and self-oscillation, in the THz device implementations.

Recent development of a new 2D material graphene necessitates sample characterization (in particular localization and distribution of defects). The presence of defects is unavoidable, however, it is possible to determine and predict defect distribution in graphene samples prior to the actual device making. A ramification algorithm is used for the above purpose.

Carbon nanotubes and graphene are promising for diverse terahertz (THz) device applications. Here, we summarize our recent studies on the THz dynamic conductivities and optoelectronic devices of these materials in the THz region. We show that the THz response of single-wall carbon nanotubes (SWCNTs) is dominated by plasmon oscillations along the nanotubes, which lead to extremely anisotropic THz conductivities. By utilizing the strong THz plasmon resonance as well as its pronounced anisotropy in aligned SWCNT films, we built THz polarizers with perfect performance and polarization-sensitive THz detectors that work at room temperature. In addition, we studied the THz conductivities of graphene samples with and without electrical gating. We demonstrated excitation and active control of surface plasmon polaritons in graphene, as well as a graphene THz modulator with a high modulation depth, a high modulation speed, and a designable resonance frequency.

The task of small target detection is to extract the small targets from the background image including clutter, noise and jitter background, so it is difficult to deal with. In this paper, after analyzing infrared small targets, noise and clutter model, we use a small window median filter to estimate the infrared background. Then using background cancelling method, that is, subtracting the estimated background from the source image, the resident image can be obtained. Finally, an adaptive threshold is used to segment the residual image to obtain the potential targets. Considering the computational load, the two-dimensional filter is simplified into a one-dimensional filter. Experimental results show that the algorithm achieved good performance and satisfy the requirement of real-time processing of large size infrared image.

This paper presents a novel image feature extraction algorithm based on multiple ant colonies cooperation. Firstly, a low resolution version of the input image is created using Gaussian pyramid algorithm, and two ant colonies are spread on the source image and low resolution image respectively. The ant colony on the low resolution image uses phase congruency as its inspiration information, while the ant colony on the source image uses gradient magnitude as its inspiration information. These two ant colonies cooperate to extract salient image features through sharing a same pheromone matrix. After the optimization process, image features are detected based on thresholding the pheromone matrix. Since gradient magnitude and phase congruency of the input image are used as inspiration information of the ant colonies, our algorithm shows higher intelligence and is capable of acquiring more complete and meaningful image features than other simpler edge detectors.

Recently, pan-tilt-zoom(PTZ) camera is widely used in extensive-area surveillance applications. A number of background modeling methods have been proposed within existing object detection and tracking systems. However, conventional background modeling methods for PTZ camera have difficulties in covering extensive field of view(FOV). This paper presents a novel object tracking system based on a spherical background model for PTZ camera. The proposed system has two components: The first one is the spherical Gaussian mixture model(S-GMM) that learns background for all the view angles in the PTZ camera. Also, Gaussian parameters in each pixel in the S-GMM are learned and updated. The second one is object tracking system with foreground detection using the S-GMM in real-time. The proposed system is suitable to cover wide FOV compared to a conventional background modeling system for PTZ camera, and is able to exactly track moving objects. We demonstrate the advantages of the proposed S-GMM for object tracking system using PTZ camera. Also, we expect to build a more advanced surveillance applications via the proposed system.

IR Target detection is one of the key technologies in military applications. However, IR sensor has limitations of passive sensor such as low detection capability to weather and atmospheric effects. In recent years, sensor fusion is active research topic to overcome the limitations. Additional active SAR sensor is selected for sensor fusion because SAR sensor is robust to various weather conditions. The state-of-the-art detector, BMVT, has good performance in clear environment such as sky and sea background for small target. However, it shows poor performance when the target has extended size or the target is located in complex background such as ground-background with dense clutters. Therefore, we presents an improved ground target detection method based on the BMVT and Morphology filter (BMVT-M). The proposed algorithm consists of two parts: The first part is target enhancement based on the BMVT. The second part is clutter rejection and target enhancement based on the Morphology filter. In addition, conventional BMVT is not suitable to SAR image for target detection because SAR image has many shot noises. Therefore we apply a median filter before the BMVT in SAR image to suppress the shot noise. For the verification of the performance, experiments are performed in various cluttered backgrounds, such as ground, sea, and sky generated by the OKTAL-SE tool. The proposed algorithm showed upgraded detection performance than the BMVT in terms of detection rate and false alarm rate. Moreover, we discuss the applicability of the proposed method to the SAR and IR sensor fusion research.