Detection and Localization of Small Moving Objects in the Presence of Sensor and Platform Movement.

In this paper, we address the challenge of detecting small moving targets in dynamic environments characterized by the concurrent movement of both platform and sensor. In such cases, simple image-based frame registration and optical flow analysis cannot be used to detect moving targets. To tackle this, it is necessary to use sensor and platform meta-data in addition to image analysis for temporal and spatial anomaly detection. To this end, we investigate techniques that utilize inertial data to enhance frame-to-frame registration, consistently yielding improved detection outcomes when compared against purely feature-based techniques. For cases where image registration is not possible even with metadata, we propose single-frame spatial anomaly detection and then estimate the range to the target using the platform velocity. The behavior of the estimated range over time helps us to discern targets from clutter. Finally, we show that a KNN classifier can be used to further reduce the false alarm rate without a significant reduction in detection performance. The proposed strategies offer a robust solution for the detection of moving targets in dynamically challenging settings.

Three-dimensional integral imaging and object detection using long-wave infrared imaging.

We propose a passive three-dimensional (3D) imaging technique based on integral imaging using a long-wave infrared (LWIR) camera. 3D imaging can improve visualization and detection of objects in adverse environments, such as low light levels and the presence of partial occlusions, along with depth estimation by reconstructing the scene at the plane of the object. This is achieved by capturing multiple two-dimensional images, known as elemental images (EI), of a scene with each image having a unique perspective of the 3D objects. Moreover, LWIR imaging performs well in photon-limited environments due to detection of thermal radiation from an object rather than the reflected light. Once the EIs have been captured, image restoration is performed on the captured images. A 3D scene is then reconstructed and object detection using correlation filters and support vector machines is performed. Our experiments with human face detection show that 2D imaging may fail to detect occluded humans, whereas passive 3D imaging with LWIR could be successful. To the best of our knowledge, this is the first report of passive 3D integral imaging with LWIR for object detection, and in particular, in low light environments.

Optimizing measurements for feature-specific compressive sensing.

While the theory of compressive sensing has been very well investigated in the literature, comparatively little attention has been given to the issues that arise when compressive measurements are made in hardware. For instance, compressive measurements are always corrupted by detector noise. Further, the number of photons available is the same whether a conventional image is sensed or multiple coded measurements are made in the same interval of time. Thus it is essential that the effects of noise and the constraint on the number of photons must be taken into account in the analysis, design, and implementation of a compressive imager. In this paper, we present a methodology for designing a set of measurement kernels (or masks) that satisfy the photon constraint and are optimum for making measurements that minimize the reconstruction error in the presence of noise. Our approach finds the masks one at a time, by determining the vector that yields the best possible measurement for reducing the reconstruction error. The subspace represented by the optimized mask is removed from the signal space, and the process is repeated to find the next best measurement. Results of simulations are presented that show that the optimum masks always outperform reconstructions based on traditional feature measurements (such as principle components), and are also better than the conventional images in high noise conditions.

Maximum Margin Correlation Filter: a new approach for localization and classification.

Support vector machine (SVM) classifiers are popular in many computer vision tasks. In most of them, the SVM classifier assumes that the object to be classified is centered in the query image, which might not always be valid, e.g., when locating and classifying a particular class of vehicles in a large scene. In this paper, we introduce a new classifier called Maximum Margin Correlation Filter (MMCF), which, while exhibiting the good generalization capabilities of SVM classifiers, is also capable of localizing objects of interest, thereby avoiding the need for image centering as is usually required in SVM classifiers. In other words, MMCF can simultaneously localize and classify objects of interest. We test the efficacy of the proposed classifier on three different tasks: vehicle recognition, eye localization, and face classification. We demonstrate that MMCF outperforms SVM classifiers as well as well known correlation filters.

3D passive integral imaging using compressive sensing.

Passive 3D sensing using integral imaging techniques has been well studied in the literature. It has been shown that a scene can be reconstructed at various depths using several 2D elemental images. This provides the ability to reconstruct objects in the presence of occlusions, and passively estimate their 3D profile. However, high resolution 2D elemental images are required for high quality 3D reconstruction. Compressive Sensing (CS) provides a way to dramatically reduce the amount of data that needs to be collected to form the elemental images, which in turn can reduce the storage and bandwidth requirements. In this paper, we explore the effects of CS in acquisition of the elemental images, and ultimately on passive 3D scene reconstruction and object recognition. Our experiments show that the performance of passive 3D sensing systems remains robust even when elemental images are recovered from very few compressive measurements.

3D passive photon counting automatic target recognition using advanced correlation filters.

In this Letter, we present results for detecting and recognizing 3D objects in photon counting images using integral imaging with maximum average correlation height filters. We show that even under photon starved conditions objects may be automatically recognized in passively sensed 3D images using advanced correlation filters. We show that the proposed filter synthesized with ideal training images can detect and recognize a 3D object in photon counting images, even in the presence of occlusions and obscuration.

Expanding the field of view by polarization multiplexing.

We introduce and demonstrate a method for expanding the field of view of a typical imaging system by multiplexing images encoded onto different polarization states and recovering them from a limited number of measurements.

Off-axis sparse aperture imaging using phase optimization techniques for application in wide-area imaging systems.

Sparse apertures find imaging applications in diverse fields such as astronomy and medicine. We are motivated by the design of a wide-area imaging system where sparse apertures can be used to construct novel and efficient optical designs. Specifically, we investigate the use of sparse apertures for off-axis imaging at infrared wavelengths while combating the effects of chromaticity to preserve resolution. In principle, several such sparse apertures can be interleaved within a common aperture to simultaneously image in multiple directions. This can ultimately lead to the design of wide-area imaging systems that require considerably less optical and electronic hardware. The resolution achievable using a sparse aperture is the same as that of a fully open aperture. In the case of off-axis imaging, however, the point spread function (PSF) introduces a blur due to chromaticity that degrades the resolution of the system. Of course, the blur can be eliminated by imaging at a single wavelength. However the signal-to-noise ratio (SNR) is poor, which ultimately degrades image quality. To improve SNR, it is necessary to widen the band of wavelengths, which of course degrades resolution due to chromaticity. Hence there is a fundamental trade between the SNR and the resolution as a function of bandwidth. We show that by using a combination of microprisms and phase optimized micropistons it is possible to reduce the chromatic blur over a band of wavelengths and improve the PSF considerably to restore the resolution of the image. The concepts are validated by means of simulations and verified with experimental data to demonstrate the advantages of phase optimized micropistons in off-axis sparse aperture imaging systems.

Automatic target recognition employing signal compression.

Quadratic correlation filters (QCFs) have been used successfully to detect and recognize targets embedded in background clutter. Recently, a QCF called the Rayleigh quotient quadratic correlation filter (RQQCF) was formulated for automatic target recognition (ATR) in IR imagery. Using training images from target and clutter classes, the RQQCF explicitly maximized a class separation metric. What we believe to be a novel approach is presented for ATR that synthesizes the RQQCF using compressed images. The proposed approach considerably reduces the computational complexity and storage requirements while retaining the high recognition accuracy of the original RQQCF technique. The advantages of the proposed scheme are illustrated using sample results obtained from experiments on IR imagery.

Integrated approach for automatic target recognition using a network of collaborative sensors.

We introduce what is believed to be a novel concept by which several sensors with automatic target recognition (ATR) capability collaborate to recognize objects. Such an approach would be suitable for netted systems in which the sensors and platforms can coordinate to optimize end-to-end performance. We use correlation filtering techniques to facilitate the development of the concept, although other ATR algorithms may be easily substituted. Essentially, a self-configuring geometry of netted platforms is proposed that positions the sensors optimally with respect to each other, and takes into account the interactions among the sensor, the recognition algorithms, and the classes of the objects to be recognized. We show how such a paradigm optimizes overall performance, and illustrate the collaborative ATR scheme for recognizing targets in synthetic aperture radar imagery by using viewing position as a sensor parameter.

Target-adaptive polarimetric synthetic aperture radar target discrimination using maximum average correlation height filters.

We report the development of a technique for adaptive selection of polarization ellipse tilt and ellipticity angles such that the target separation from clutter is maximized. From the radar scattering matrix [S] and its complex components, in phase and quadrature phase, the elements of the Mueller matrix are obtained. Then, by means of polarization synthesis, the radar cross section of the radar scatters are obtained at different transmitting and receiving polarization states. By designing a maximum average correlation height filter, we derive a target versus clutter distance measure as a function of four transmit and receive polarization state angles. The results of applying this method on real synthetic aperture radar imagery indicate a set of four transmit and receive angles that lead to maximum target versus clutter discrimination. These optimum angles are different for different targets. Hence, by adaptive control of the state of polarization of polarimetric radar, one can noticeably improve the discrimination of targets from clutter.

Quadratic correlation filter design methodology for target detection and surveillance applications.

A novel method is presented for optimization of quadratic correlation filters (QCFs) for shift-invariant target detection in imagery. The QCFs are quadratic classifiers that operate directly on the image data without feature extraction or segmentation. In this sense, the QCFs retain the main advantages of conventional linear correlation filters while offering significant improvements in other respects. For example, multiple correlators work in parallel to optimize jointly the QCF performance metric and produce a single combined output, which leads to considerable simplification of the postprocessing scheme. In addition, QCFs also yield better performance than their linear counterparts for comparable throughput requirements. The primary application considered is target detection in infrared imagery for surveillance applications. In the current approach, the class-separation metric is formulated as a Rayleigh quotient that is maximized by the QCF solution. It is shown that the proposed method results in considerable improvement in performance compared with a previously reported QCF design approach and many other detection techniques. The results of independent tests and evaluations at the U.S. Army's Night Vision Laboratory are also presented.

Constrained quadratic correlation filters for target detection.

A method for designing and implementing quadratic correlation filters (QCFs) for shift-invariant target detection in imagery is presented. The QCFs are quadratic classifiers that operate directly on the image data without feature extraction or segmentation. In this sense the QCFs retain the main advantages of conventional linear correlation filters while offering significant improvements in other respects. Not only is more processing required for detection of peaks in the outputs of multiple linear filters but choosing the most suitable among them is an error-prone task. All channels in a QCF work together to optimize the same performance metric and to produce a combined output that leads to considerable simplification of the postprocessing scheme. The QCFs that are developed involve hard constraints on the output of the filter. Inasmuch as this design methodology is indicative of the synthetic discriminant function (SDF) approach for linear filters, the filters that we develop here are referred to as quadratic SDFs (QSDFs). Two methods for designing QSDFs are presented, an efficient architecture for achieving them is discussed, and results from the Moving and Stationary Target Acquisition and Recognition synthetic aperture radar data set are presented.

Biometric verification with correlation filters.

Using biometrics for subject verification can significantly improve security over that of approaches based on passwords and personal identification numbers, both of which people tend to lose or forget. In biometric verification the system tries to match an input biometric (such as a fingerprint, face image, or iris image) to a stored biometric template. Thus correlation filter techniques are attractive candidates for the matching precision needed in biometric verification. In particular, advanced correlation filters, such as synthetic discriminant function filters, can offer very good matching performance in the presence of variability in these biometric images (e.g., facial expressions, illumination changes, etc.). We investigate the performance of advanced correlation filters for face, fingerprint, and iris biometric verification.

Object recognition in subband transform-compressed images by use of correlation filters.

We introduce subband correlation filters (SCFs) as a solution to the problem of object recognition at multiple resolution levels in quantized transformed imagery. The approach synthesizes correlation filters that operate directly on subband coefficients rather than on image data. We explore two techniques to accomplish the reduced-resolution recognition: (1) training the correlation filters to incorporate downsampling tolerance and (2) adaptation of the subband decomposition filters to accommodate the reduced resolutions. For compression ratios of 20:1, SCFs demonstrate recognition performance of at least 90%, 85%, and 75%, respectively, on 2-, 4-, and 8-ft-resolution synthetic aperture radar data.