Proceedings Volume 2234

Automatic Object Recognition IV

cover
Proceedings Volume 2234

Automatic Object Recognition IV

View the digital version of this volume at SPIE Digital Libarary.

Volume Details

Date Published: 29 July 1994
Contents: 4 Sessions, 41 Papers, 0 Presentations
Conference: SPIE's International Symposium on Optical Engineering and Photonics in Aerospace Sensing 1994
Volume Number: 2234

Table of Contents

icon_mobile_dropdown

Table of Contents

All links to SPIE Proceedings will open in the SPIE Digital Library. external link icon
View Session icon_mobile_dropdown
  • Multisensor Processing for Automatic Object Recognition
  • Radar Processing for Automatic Object Recognition
  • Sonar Processing for Automatic Object Recognition
  • Multisensor Processing for Automatic Object Recognition
  • FLIR Processing for Automatic Object Recognition and Novel Techniques
  • Radar Processing for Automatic Object Recognition
  • Sonar Processing for Automatic Object Recognition
Multisensor Processing for Automatic Object Recognition
icon_mobile_dropdown
Model-based 3D object recognition using intensity and range images
Emerico Natonek, Charles Baur
This paper describes a model based vision system in which a commercial 3-D computer graphics system has been used for object modeling and visual clue generation. Given the computer generated model image (i.e., color, depth, ...) a conventional CCD camera image and the corresponding scanned 3-D dense range map of the real scene, the object can be located in it. Our system, called three dimensional model based approach (3D-MBA), uses image pyramid of resolution and prediction-verification processes. To optimize the object recognition scheme, it first forms a set of hypotheses about the objects present in the scene and then proceeds by trying to confirm/reject them. If any part of the object hypothesis is missing, the system uses the object model to predict the shape, location, and orientation of that missing part. This paper focuses on how this is done using newly developed segmentation algorithms extracting `regions of interest' from range images (depth map) of the scene. Illustrative examples of object recognition in simple and complex scenes are presented.
Radar Processing for Automatic Object Recognition
icon_mobile_dropdown
Aspect dependence of impulse radar extracted signature features in the combined time-frequency domain
Hans C. Strifors, B. Brusmark, Steffan Abrahamson, et al.
We study the scattering interaction of electromagnetic pulses of short duration with a target of either simple or complex shape. The complex targets are plastic scale models of two aircraft with metallized surface. The form-function in the back-scattering direction, filtered by the complete system of radar and digitizing signal analyzer, is determined from measured data when the target is illuminated at each one of several different aspect angles. The experimental data are obtained with the aid of an impulse radar system located in an anechoic chamber. We examine the details of how various resonance features of the returned echo evolve in time using a PWD with a suitably selected Gaussian time-window. The results are displayed in a series of 3-D surface plots and 2-D contour plots together with the recorded waveforms and their spectra. Time-frequency signatures from the two scale models of aircraft are examined.
Fully polarimetric generalized likelihood ratio tests (GLRTs) for detecting scattering centers with unknown amplitude, phase, and tilt angle in terrain clutter
We present a family of polarimetric generalized likelihood ratio tests (PGLRTs) which exploit fully polarimetric information in a high resolution application to detect scattering centers in terrain clutter. The detectors are based on a deterministic target model derived from the Huynen parameterization of a scattering matrix. The model is parameterized by target amplitude, absolute phase, and target orientation angle. These parameters, which are unknown in many practical applications, are estimated by the detectors. the PGLRTs may be used to enhance the responses of certain scattering center types relative to others in a given region of interest. Once a scattering center is detected, the ML estimates formed by a PGLRT may be used to further describe the detected target. We implement and analyze the performance of the PGLRTs designed for Gaussian and K-distributed clutter with known covariance. The PGLRT that assumes all three model parameters are unknown is a detector whose performance we show to lie between that of the optimal polarimetric detector and the polarization whitening filter.
Localized radon transform for ship wake detection in SAR imagery
Anthony C. Copeland, Gopalan Ravichandran, Mohan M. Trivedi
The radon transform is commonly used in algorithms for detecting linear features in an image. However, it can have difficulty detecting line segments that are significantly shorter than the image dimensions, and has no capability of providing information about the positions of the endpoints of these shorter line segments, or on line length. These problems are magnified when the transform is applied to an image with a high level of noise. Our localization of the radon transform reduces the spatial extent of the image intensity integration, to improve the detection and localization of short line segments. This localized radon transform forms the basis of a linear feature detection algorithm that is demonstrated on several synthetic images containing various levels of random noise and on actual images containing linear features. Our experimental results demonstrate the ability of this approach to successfully detect the linear components of ship wakes visible in SAR ocean imagery.
Automatic calibration of laser range cameras using arbitrary planar surfaces
James E. Baker
Laser range cameras (LRCs) are powerful tools for many robotic/computer perception activities. An LRC's output is an array of distances obtained by scanning a laser over the scene. To accurately interpret this data, the angular definition of each pixel must be known. Typically, the range data is converted to Cartesian coordinates by calibration-parameterized, non-linear transformation equations. This paper presents an automated method which uses genetic algorithms to search for calibration parameter values and possible transformation equations which combine to maximize the planarity of user-specified sub-regions of the image(s). This method permits calibration to be based on an arbitrary plane, without precise knowledge of the LRC's mechanical precision, intrinsic design, or its relative positioning to the target. Furthermore, this method permits rapid, remote, and on-line recalibration. Empirical validation of this system has been performed using two different LRC systems and has led to significant improvement in image accuracy while reducing the calibration time by orders of magnitude.
Advances in automatic estimation and compensation of target kinematics for improved radar imaging
Three techniques for the motion compensation of wideband frequency-stepped signatures are compared. These approaches were developed with the intention of processing the signatures into focused range-Doppler images of moving targets. The approaches differ in three aspects: (1) the amount of Fourier processing; (2) the optimization process that yields the motion parameter estimates; and (3) the type of function utilized in the optimization process. Here, the emphasis is placed on the choice of a function that determines the motion parameters. The basic premise is that the function selected has a global minimum whose coordinates are the optimum estimates of the true motion parameters. In the first approach, the function is a measure of the entropy associated with the image. In the second approach, the function is a measure of the entropy associated with the range profile history from which the image is formed. In the third approach, the function is a measure of the rate of change of the target signature. A test case is offered to illustrate the properties of these functions. For this particular case, the performance of each motion compensation approach depends on the behavior of the corresponding function over a selected domain that includes the actual motion parameters of the target.
Model-based automatic target recognition from high-range-resolution radar returns
John S. Baras, Sheldon I. Wolk
We develop economic target descriptions based on high range resolution target returns, utilizing wavelet multiresolution representations and tree structured vector quantization, in its clustering mode. The algorithm automatically constructs the multi-scale aspect graph of the target. This results in a progressive coding of the target model information and in an extremely efficient, hierarchical indexing of the stored target models. As a final outcome we obtain extremely fast recovery, search, and matching during the on-line ATR operation. We also investigate, the so-called new target insertion problem in a fielded ATR system, and the required fast reprogrammability of the ATR system. We compare the performance and cost (both computational and hardware) of ATR algorithms based on the parallel use of single target aspect graphs vs ATR algorithms using the combined aspect graph for the group of targets under consideration. We show that efficient real-time ATR algorithms can be constructed using the aspect graph of each target in a parallel computation. The resulting architecture includes wavelet preprocessing with neural networks postprocessing. We use synthetic radar returns from ships as the experimental data to demonstrate the performance of the resulting ATR algorithm.
Sonar Processing for Automatic Object Recognition
icon_mobile_dropdown
Characteristic resonance signatures for acoustical scattering from fish
Christopher Feuillade, R. H. Love, Michael F. Werby
Scattering from fish is an important issue for biologists, aquaculturalists, and the military community. One can track the habits of fish, determine their abundance, and learn to distinguish them from targets of military interest by accurate mathematical formulations of the problem. The aim of this work is to present a mathematical/physical formulation by coupling acoustical signals with fish bladder properties to describe the acoustical scattering from fish bladders. In this work we first develop the theory to describe acoustical scattering from a viscous spherical shell filled with air; and then use it to show how the scattering amplitude and Q of the monopole resonance are modified by the viscosity and thickness of the shell. We then show how the theory may be extended, using a mathematical formulation based on coupled boundary integral equations, to account for the systematics of fish bladder scattering when elongation and viscous effects are combined.
Pulse scattering from objects near an ocean surface and the inverse problem of target identification
Elmer White, Michael F. Werby
We have developed the correct theory to describe scattering from targets near an interface. Pulse signals are particularly interesting because with proper placement of source and receiver it is possible to extract information about target depth and target characteristics from resonance properties. In this work we present the correct formulation of scattering from a target near an interface, and we present form functions and pulse return signals for spheroidal shells near an ocean surface or rigid bottom. Results for several cases are described.
Broadband pulse signals in an ocean waveguide as an inverse tool for the rapid extraction of some ocean properties
Michael F. Werby, Elmer White
With the inception of a new and fast normal-mode model that accurately simulates signals propagating in an ocean waveguide, it is possible to reconstruct pulse signals in the ocean. By measuring the pulse signals at different depths, one can extract a variety of information about the waveguide, such as ducting properties associated with the velocity profile, and some bottom properties that relate to the pulse widths. We present examples of the extraction data from this methodology.
Pulse returns from elongated elastic shells and the extraction of target characteristics
Cleon E. Dean, Michael F. Werby
It is now possible to model elongated elastic targets via the Waterman methodology of the extended boundary condition equations. Due to rapid advances in computers it is now possible to perform numerical calculations of the scattered field in any direction as a function of frequency. We take advantage of this capability to study pulse signals from such targets in which we vary shell material, shell thickness, and elongation of the target. The pulse signals are then reconstructed via Fourier transform methods and the results analyzed and interpreted.
Identification of cylindrical shells with sets of reinforcing ribs by their acoustic Bragg diffraction (grating) patterns
Donald Brill, Guillermo C. Gaunaurd
We present a basic study of the acoustic Bragg diffraction patterns produced by gratings with various types of slits, which are insonified by (monochromatic) acoustic waves of various frequencies. Given the spacing in-between the slits, the (Bragg) angular diffraction pattern generated by the grating can be predicted by the method of physical optics (acoustics), and can be numerically evaluated for gratings with either uniformly or irregularly spaced slits. It has been already established that the diffraction patterns of planar gratings can be used to model the far-field scattering patterns produced by thin, cylindrical, elastic shells in water that have many reinforcing ribs. Varying the size and slit spacing (i.e., the ribs), the diffraction pattern becomes distorted, and for very irregularly spaced cases, the distortion is substantially asymmetric and noticeable. A number of graphs are shown to illustrate these points.
Holographic analysis of the dynamics and structure of sonar transducer arrays
The discrimination and recognition of underwater targets by an operating sonar array depends on its true structural dynamics. These dynamics are directly related to the mechanical and physical interfaces between transducers and their supporting and radiating structures. Such complex assemblies require sophisticated bonding and processing of compliant and rigid materials which determine the operational parameters of the sonar system. Understanding the effects of structural anomalies and defects is of critical importance to insuring correct array operation and a nondestructive method for the visualization of anomalous structural characteristics would be of great value in both development and testing of sonar systems. Knowledge of the mechanisms for acoustic energy transfer through the sonar structure into the water medium can be inferred from these visualizations and would enhance the analysis of returned signals for homing and target recognition in operation. Holographic interferometry has presented itself as a viable and useful method for the realization of this type of information.
Remote sensing of moving sources with complex spatial-time structure
Natalia A. Sidorovskaia, Iosif Sh. Fiks, Victor I. Turchin
The theoretical concept and algorithms of experimental signal processing, permitting the determination of complex acoustical source characteristics from near-field data, are discussed. The algorithms are based on two different approaches: on the integral transformations using a high-frequency approximation and on the methods of parametric estimation theory. The theoretical part of near-field methods includes the design of algorithms of measured data transformation, the investigation of the angular sectors of trustworthy determination of the direction pattern and resolution of image reconstruction, and the analysis of requirements for the receiving system. The numerical simulation of the direction pattern reconstruction and experimental result of the reconstruction of the ship acoustical images in the frequency-spatial domain from linear to array data are included.
Sound scattering by spatial-localized inhomogeneities in oceanic waveguides: calculation and measurement methods
Sergey M. Gorsky, Vitaly A. Zverev, Alexander I. Khil'ko
Diffraction of the acoustic fields by spatially localized inhomogeneities in oceanic waveguides is investigated by analytical methods using numerical experiments on the basis of physical modeling and by way of field measurements. The possibilities of numerical simulation and measurement under the conditions of physical modeling are discussed. It was shown that from the point of view of studying the diffraction phenomena in the ocean, the significance of these methods is due to the difficulties in performing hydroacoustic experiments. Provisional calculations and measurements under the model conditions can noticeably increase the efficiency of sophisticated field observations. Besides using the results obtained by different authors, a systematic concept of the structure of perturbed signals in the waveguides and a brief analysis of the possibility of the formulation and measurement of diffracted fields in layered waveguides are proposed.
Multisensor Processing for Automatic Object Recognition
icon_mobile_dropdown
Vision planner for an intelligent multisensory vision system
Xiaoyi Jiang, Horst Bunke
In this paper we present a multisensory vision system that is intended to support the vision requirements of an intelligent robot system. Contrary to many other vision systems, our system has two significant new features. First, it contains multiple sensors, object representations, and image analysis and interpretation methods in order to solve a number of different vision tasks. Secondly, it comprises a vision planner. Upon a task-level vision request from the robot system, the vision planner transforms it into an appropriate sequence of concrete vision operations, executes these operations, and if necessary, finds out alternative sequences of operations. Experimental results demonstrate the clear advantage of this combination of multiple resources with the vision planner in solving typical vision problems for robotic tasks.
Automated training of 3D morphology algorithm for object recognition
Michael E. Bullock, David L. Wang, Scott R. Fairchild, et al.
Grayscale morphology has demonstrated a great deal of success in automatic target recognition (ATR) applications with a variety of imagery sources including SAR, IR, visible, and multispectral. However, training the morphology algorithm requires significant experience and is labor intensive. This paper presents an innovative approach for using genetic algorithms (GA) and the classification and regression trees (CART) algorithm to automate morphology algorithm training and optimize detection performance. The GA is used to find the morphology operators by encoding them into binary vectors. The CART algorithm determines the optimum region filtering parameters in conjunction with the morphology operations. Robustness is achieved by regression pruning of the CART generated classification trees. The basic concepts in applying the GA to the design of grayscale morphology filters is described. Our results suggest that the detection performance of a GA designed morphology filter is comparable to that designed by human experts. The automated design method significantly shortens the design process.
Radiometrically correct sharpening of multispectral images using a panchromatic image
Tim J. Patterson, Michael E. Bullock
With the advent of a panchromatic band on the enhanced thematic mapper (ETM) sensor, the interest in sharpening techniques for multispectral imagery has substantially increased. Previous work has been performed to increase the spatial resolution of multispectral imagery by combining it with a higher spatial resolution broad band panchromatic source. While these techniques resulted in a sharper multispectral image, they limited the extent to which the spectral information could be used. In this paper, we describe a technique that employs pseudoinverse estimation to generate a sharpened multispectral image. This technique provides the minimum mean squared error estimate of the sharpened spectral signatures given the information available from the panchromatic and multispectral sensors. The resulting radiometrically correct sharpened image is useful for various types of spectral analysis such as spectral classification, demixing, and detection. We then present the results of a quantitative evaluation of the processing technique in various operating modes as well as the results of a qualitative evaluation from analysts using MSI for mapping purposes.
Hybrid optical system for frequency-domain texture classification of imagery
Roger L. Easton Jr., Marc J. Comeau, Scott D. Brown, et al.
An algorithm for segmenting images based on local texture properties and which requires only linear filtering operations has been described in a previous paper. We now report on the implementation of this algorithm in a hybrid electro-optical system which performs the linear filtering operations in the spatial frequency domain. Liquid-crystal video displays (LCDs) are used for input of the image and the frequency-domain filter. The system has the potential for rapid segmentation of simple images by texture properties.
Polarization-sensitive thermal imaging
Cornell S. L. Chun, David L. Fleming, E. J. Torok
In order to recognize 3-D objects, conventional methods in robot vision perform shape extraction by sensing the intensity of light reflected by objects. A fundamental problem associated with sensing the intensity of reflected light is that intensity gives one parameter while the surface orientation of objects has two degrees of freedom. Physics Innovations Inc. is developing a thermal imaging technique for determining surface orientation where, in each image pixel, two parameters are sensed simultaneously. The two parameters, percent of polarization P and angle of the plane of polarization (phi) , are directly related to the two angles of surface orientation. In this paper the uncertainties in determining P, and (phi) using the Physics Innovations sensor are made explicit by analytical expressions and by computer simulations of the images. These uncertainties are related to temporal and spatial noise characteristics of the imaging system and to the polarization efficiencies of the polarizers. Automatic object recognition using polarization information is dependent on the uncertainties in determining P, and (phi) .
Random-set approach to data fusion
This paper describes a fundamentally new theoretical approach to data fusion based on a novel type of random variable called the random finite set, and on a generalization of the familiar radon-nikodym derivative from the theory of the Lebesgue integral. We have shown how to directly generalize classical (i.e., single-sensor, single-target) parametric point estimation theory to the multi-sensor, multi-target, localization and classification realm. Using this theory we have shown that it is possible to construct data fusion algorithms in which detection, correlation, tracking and classification are unified into a single probabilistic procedure. We have also shown that a Cramer-Rao inequality holds for a general class of data fusion algorithms, apparently the first ever.
Classification when a priori evidence is ambiguous
This paper describes conditioned Dempster-Shafer (CDS) theory, a probabilistic calculus for dealing with possibly non-Bayesian evidence when underlying a priori knowledge is possibly non-Bayesian. The Dempster-Shafer composition operator can be `conditioned' to reflect the influence of any kind of a priori knowledge which can be modeled as a Dempster-Shafer belief measure. CDS is firmly grounded in probability theory via the theory of random sets. It is also a generalization of the Bayesian theory of evidence to the case when both evidence and a priori knowledge are ambiguous.
Distributed object recognition using fuzzy relational inference logic
Phil Greenway
This paper describes a system which uses fuzzy relational inference logic (an implementation of Dempster-Schafer evidential reasoning) to combine identity estimates derived from a network of distributed sensing nodes. The temporal association is mediated through the use of a multi-target tracking system built around a decentralized Kalman filter, and different combination rules are applied for the cases of consistent or conflicting evidence. Comparisons are drawn with approaches based on the explicit computation of identity probability estimates and their combination. The availability of good estimates of target identity can be used to resolve some of the basic data association ambiguities in the multi-target tracking system. This paper reviews some background material in data fusion. Then describes the vision component of the decentralized data fusion test-bed which has been used as the basis for the system considered here. The basic identity fusion algorithm is then presented, and a comparison drawn with an alternative, Bayesian approach. The possible extension of the system to include neural network based target classification is also considered.
FLIR Processing for Automatic Object Recognition and Novel Techniques
icon_mobile_dropdown
Automatic target recognition during sensor motion and vibration
Ofer Hadar, A. Kuntsevitsky, M. Wasserblat, et al.
The resolution capability of imaging systems is effected by a blurring effect due to the vibration and motion recorded in the image. This disturbance is often more severe than electronic and optical limitations inherent in the system. This fact must be considered when dealing with the development and analysis of automatic target recognition (ATR) systems for military applications. The aim of this research is to analyze the influence of image vibrations and motion upon the probability of acquiring the target with an ATR system. The analysis includes factors that characterize the relationship existing between the target and its background. A high level of correlation is expected between these factors and the probability of target detection, enabling efficient performance in the prediction and evaluation of any ATR system. The results of this research can be implemented in military applications as well as in developing image restoration procedures for image-blur conditions.
Multiple target detection using fringe-adjusted joint transform correlator
In this paper, we present a novel technique for multiple target detection using fringe-adjusted joint transform correlator. This technique, involving adjustment of the joint power spectrum (JPS) on the basis of the input-scene-only power spectrum, and apodization of the modified JPS on the basis of the reference signal power spectrum, is found to yield good correlation output. Simulation results are presented to verify the performance of the proposed technique.
Self-partitioning neural networks for target recognition
Heggere S. Ranganath, S. Richard F. Sims
In this paper, we present a method for quantifying the degree of non-cooperation that exists among the target members of the neural network training set. Both the network architecture and the training algorithm are taken into consideration while computing non-cooperation measures. Based on these measures the network automatically partitions into several identical networks and each partition learns a subset of the targets. The partitioning takes place only when necessary and when needed the computation for partitioning is minimal. Each network is simple with only one hidden layer and currently has only one node in the output layer. A fusion network combines partial results to produce the final response. Simulation results indicate that the method is robust and capable of self organization to overcome the ill effects of non-cooperating targets in the training set, thereby reducing training time significantly.
Feature assessment in imperfectly supervised environments
This study extends the previously reported feature assessment scheme CORPS (class overlap region partitioning scheme) designed for perfectly supervised environments to the imperfectly supervised domain. The imperfectness levels of the labels, which can be different for different classes, are used to appropriately weight the feature space overlap evaluation process, i.e., the samples from classes with more reliable labels are given correspondingly more weightage than those with less reliable labels. The methodology can be applied to mixed supervised and imperfectly supervised environments also, with subsets of data even within a class having different imperfectness levels. Like CORPS, the extended method can be used either as a stand alone tool or as a front end to more complex combinatorial feature selection procedures such as branch and bound and genetic algorithms. The new approach also has the flexibility to permit a bias in favor of either of the two possible types of errors in a binary decision process, such as false alarm and leakage in a target detection problem. Algorithmic and operational details are included to facilitate wide usage of this new tool.
Adaptive learning concepts and methodology for enhanced recognition system performance
This study presents innovative adaptive learning concepts and resulting methodology designed for enhancing the performance of recognition systems such as automatic target recognition (ATR) systems wherein robustness of performance is a significant issue. The basic underlying concept is that of learning in partially exposed environments, wherein the system is not necessarily aware of all the pattern classes that may be encountered in the operational phase. The methodology is based on such learning as required by nearest neighbor based decision systems. The paper discusses several stages of sophistication of the system design and illustrates these with two sets of numerical experiments (using the now classical iris data as well as some real-world TV image data) to bring out the subtleties of the issues involved.
Adaptive correlation-based tracking algorithm
Mark W. Eklund, Mohan M. Trivedi
The tracking algorithm introduced in this paper was designed with the intent of operation on a pipeline processor. This is different from other algorithms that rely on the branching capabilities which can easily be accomplished by a conventional processor but not pipeline processors. The localization of an object from one frame and the following frame is done using correlation which a pipeline processor handles very well. Finally, little prior knowledge of the object being tracked is required. The heart of this algorithm is the correlation mask. The mask used is a balance between the popular normalized cross correlation filter and the MACE filter. The masks are created from the actual image. Since the object is not a static object, the masks are updated while the object is being tracked. This paper discusses the implementation of this algorithm and its testing.
Real-time signature verification using neural network algorithms to process optically extracted features
Ned Francis O'Brien, Steven C. Gustafson
A technique for handwritten signature verification is proposed which combines the pattern recognition abilities of neural networks with the feature extraction capabilities of optics. This two-part technique enables real time signature verification based upon power spectrum features and stored linear least squares and Gaussian radial basis function neural network weights.
Minimum average correlation energy (MACE) prefilter networks for automatic target recognition
Gregory L. Hobson, S. Richard F. Sims, Paul D. Gader, et al.
Minimum average correlation energy (MACE) filters have been shown to be an effective generalization of the synthetic discriminant function (SDF) approach to automatic target recognition. The MACE filter has the advantage of having a very low false alarm rate, but suffers from a diminished probability of detection. Several generalizations have recently been proposed to maintain the low false alarm rate while increasing the probability of detection. The mathematical formulation of the MACE filter can be decomposed into a linear `prefilter' followed by an SDF-like operation. It is the prefiltering portion of the MACE which accounts for the low false alarm rate. In this paper, we insert a nonlinearity in this process by replacing the SDF portion of the operation by a neural network and show that we can increase the probability of detection without sacrificing low false alarm rates. This approach is demonstrated on a standard multiaspect image set and compared to the MACE and its generalizations.
High-performance real-time classification processor using conventional and neural processing techniques
Robert Shears
Future systems must intelligently preprocess an increasingly large volume of sensor data in real time within the constraints of an airborne environment. A compact, flightworthy, real time sensor processor has been developed which can host both conventional and neural processing algorithms. This will support a wide range of applications including high performance target classification from high resolution infrared imagery. The architecture features software reconfigurability to support different processing tasks and modular upgrade potential. Accurate classification of ship imagery has been demonstrated, using real airborne trials data produced under a range of operating conditions. The system is currently continuing to undergo airborne trials.
Jump-diffusion processes for the automated understanding of FLIR scenes
Aaron D. Lanterman, Michael I. Miller, Donald L. Snyder, et al.
We take a pattern theoretic approach to recognizing and tracking ground-based targets in sequences of forward-looking infrared images acquired from an airborne platform. A rich set of transformations on objects represented by 3D faceted models are formulated to accommodate the variability found in FLIR imagery. An hypothesized scene, simulated from the emissive characteristics of the hypothesized scene elements, is compared with the collected data by a likelihood function based on sensor statistics. This likelihood is combined with a prior distribution defined over the set of possible scenes to form a posterior distribution. A jump-diffusion process empirically generates the posterior distribution. The jumps accommodate the discrete aspects of the estimation problem, such as adding and removing hypothesized targets and changing target types. Between jumps, a diffusion process refines the hypothesis by following the gradient of the posterior. Since the likelihood function may include likelihoods from other sensors and may be defined over past and current times, interframe processing and sensor fusion are natural consequences of the pattern theoretic approach.
Application of genetic algorithm for automatic recognition of partially occluded objects
Automatic recognition of partially occluded objects that are sensed by imaging sensors is a challenging problem in image understanding (IU), automatic target recognition (ATR), and computer vision fields. In this paper I address this problem by using a genetic algorithm (GA) as part of a model-based recognition scheme. The partially occluded object segments are rotated, translated, and scaled. Then each transform parameter is encoded into a binary string and used in a genetic algorithm. The suggested transformation is then applied to the sensed segment and the resulting object is matched against a library of stored targets. The fitness criterion is a distance function that measures the similarity between the segmented object and the stored target models. The GA by performing the process of mutation, reproduction, and crossover suggests optimum transform parameter sets. The empirical results of the application of the approach on a set of real ladar data of military targets shows that correct recognition for up to 50% target occlusion is possible.
Radar Processing for Automatic Object Recognition
icon_mobile_dropdown
Radar target modeling: a geometric theory of diffraction (GTD)-based approach
Rob Carriere, Lee C. Potter, Randolph L. Moses, et al.
A new approach to scattering center extraction is developed based on a model derived from the geometric theory of diffraction (GTD). For stepped frequency measurements at high frequencies, this model is better matched to the physical scattering process than the Prony or discrete Fourier transform modeling methods. In addition, the GTD-based model extracts more information about the scattering centers, allowing partial identification of scattering center geometry in addition to determining energy and downrange distance. We derive expressions for the Cramer-Rao bound of this model; using these expressions we analyze the behavior of the new model as a function of scatterer separation, bandwidth, number of data points, and noise level. We compare these results with those for the Prony model. Additionally, a maximum likelihood algorithm for the model is developed. Estimation results using data measured on a compact range are presented to validate the proposed modeling procedure.
Evaluation of single- and full-polarization two-dimensional Prony techniques applied to radar data
Joseph J. Sacchini, Anthony Romano, William M. Steedly
The recently developed 2-D TLS-Prony technique is applied to a single and full-polarization synthetically generated radar data set. The radar target analyzed is a generic aircraft consisting of a fuselage, wing, stabilizer, and tail. The 2-D TLS-Prony technique is a parametric estimation technique that models the radar return frequency domain data (multiple angle, multiple frequency data, e.g., SAR/ISAR) using damped exponentials. The technique's ability to accurately model and characterize the target are investigated. Scattering center location and characterization are accomplished by the technique. Issues such as model order selection, bandwidth requirements, and modeling error are examined for both single and full polarization data sets.
Applications of wavelet subband decomposition in adaptive arrays
Pre-processing radar signals incident on an adaptive array by applying an invertible transformation such as wavelets are the focus of this study. The effect of wavelet subband decomposition of radar signals prior to adaptation using an LMS algorithm or an Applebaum processor on the adaptation rate of these processors is examined. The performance of wavelet transform based array processors is compared with that of the FFT and cosine transform. The dynamic range of the array weights before and after wavelet transformation is being examined. Simulations involving experimental radar data and different types of wavelets are also presented.
Sonar Processing for Automatic Object Recognition
icon_mobile_dropdown
Long-range cw and time-domain simulations of ocean acoustic scatter from Mid-Atlantic Ridge corners
Stanley A. Chin-Bing, David B. King, Joseph E. Murphy, et al.
Long-range acoustic experiments done in the mid-Atlantic ridge region show strong backscatter from many ocean bottom features. One type of feature characteristic of this region, and a possible contributor to this acoustic backscatter, takes the form of a seafloor ridge corner formed by two ridge faces intersecting at near-right angles. We have investigated through computer simulations the effects of an anelastic seafloor corner on the backscattered acoustic field. The acoustic source insonifying the seafloor corner was near the sea surface and approximately 4 km away in range. The ocean depth at the seafloor corner was also 4 km. Simulations were made using cw computer models. Time-domain calculations were obtained from FFTs of the cw fields. Examples are presented that show large backscatter near the base of the ridge while relatively low backscatter near the ridge peak. This is attributed to the angle of the insonifying field and multiple scattering effects. Backscatter from a longer range (10 - 20 km), gently sloped sedimented seafloor is also presented and discussed.
Scattering effects of internal waves in the shallow-water ocean environment
David B. King, Stanley A. Chin-Bing
The effects on acoustic energy caused by variations in the ocean environment have long been of interest to the ocean acoustics community. The shallow-water environment can be highly variable on both temporal and spatial scales. An example of shallow water environmental variability is the phenomena of internal waves. Under certain constraints, such as the spatial extent of the internal wave and the acoustic frequency, internal waves can produce very significant effects (on the order of 20 dB) on the propagation of acoustic energy. Depending on the geometry of the source and the target, the presence of an internal wave can either enhance or degrade the strength of a signal. In this work we (1) demonstrate the redistribution of acoustic energy that can be caused by the presence of shallow water internal waves, (2) show that this loss in propagated energy involves mode conversion coupled with a lossy shallow water bottom, and (3) illustrate that this phenomena can occur at a variety of frequencies that are dependent on the soliton's varying size and physical configuration. A significant implication is that solitons could be a possible explanation for the anomalous shallow-water propagation loss observed at a number of different frequencies.
Evaluation of the near-field impulse response of hard oblate spheroids in the limit of approaching a thin disk
Guy V. Norton, Jorge Novarini
Kristensson and Waterman developed a method utilizing a T-matrix approach to obtain the scattered field from a disk. The solution was obtained by setting up the equations for an oblate spheroid and evaluating them in the limit of vanishing minor axis. In this work the reflected and diffracted components of the scattered field are examined for the case of an oblate spheroid of large aspect ratio (minor axis small compared to major axis). To separate the reflected and diffracted contributions of the scattered field requires that the near-field impulse response be determined. The impulse response is obtained via Fourier transformation of the frequency response. The impulse response reduces each of the phenomena involved (reflection, diffraction, and secondary diffractions) to a set of discrete events in time which are manifested by a series of spikes, each with their own amplitude and temporal location. Evaluation of these events aid in determining the physical size and type of scatterer (target identification). The associated theory and numerical experiments are presented.
Simulation of broadband propagation and scattering in a waveguide
Rob A. Zingarelli
Using a SIMD-optimized broadband parabolic equation propagation model in a range- dependent waveguide, the extended boundary condition model for object scattering, and a new method for coupling the two models, broadband basin-scale active acoustic simulations are possible. Preliminary results and reference solutions generated using a finite difference time domain method are presented.