Faculty
Students
Research
Publications
Seminars

       

    NC State University
    ECE Department
    Raleigh, NC 27695

Vision, Information and Statistical  Signal Theories and Applications group                             

Workshop:

2005 Program Review for Sensing, Imaging and Object Recognition
May 25-27, 2005 NCSU, Raleigh NC, USA

Presentations Download

Venue • Program • Schedule • Events • Abstracts • Directions • Housing • Registration

Organizers: H. Krim and J. Sjogren

This program review aims to bring together researchers in the areas of sensing, imaging and object recognition to interact and share their latest results. This forum of exchange and crossfertilization of ideas will assemble engineers, computer scientists, physicists and mathematicians.

Venue:

Monteith Engineering Research Center

Completed in 1996, the 138,000-square-foot Monteith Engineering Research Center houses some of the most sophisticated research facilities on Centennial Campus. A complex of two buildings, the MRC features a wide range of state-of-the-art technology, from special “clean-rooms” for creating and analyzing nanoscale microelectronic devices to the adjoining Constructed Facilities Laboratory, which contains some of the largest testing equipment in the nation.

Originally named the Engineering Graduate Research Center, the building, designed by architects at O'Dell Associates, was renamed in 2005 to honor former Chancellor Larry K. Monteith.

The workshop will be held in Room 136

Program:

25 May 2005

1830-2100 Reception at Engineering Building I, Centennial Campus

26 May 2005

0830-0930  Plenary I -- Toward Brain and Computer Interfacing, EGRC 136, M. Kirby

0930-1000 Talk 1 -- Multiscale Geometric Signal Processing in High Dimensions, H.  Choi

1000-1030 Break: Room EGRC 246

1030-1100 Talk 2 -- Optimal probing for system identification with applications to Biomolecular NMR, N.  Khaneja  

1130-1200 Talk 3 -- Free-Space Laser Communications: Propagation and Beam Formation Investigations, M. Giles  

1200-1230 Talk 4 -- An Introduction to RHINO (Real-time Histogram Interpretation of Numerical Observations, G. Lukesh

1230-1330 Lunch: EGRC 246

1330-1400 Plenary 2 -- SENSING AND COMMUNICATIONS USING ULTRAWIDEBAND RANDOM NOISE WAVEFORMS, R. Narayanan

1400-1430 Talk 5 -- SHARPENING TECHNIQUES FOR SENSOR FEATURE ENHANCEMENT, L.  Marple

1430-1500 Talk 6 –TBA Zetterlind

1500-1530 Break: EGRC 246

1530-1600 Talk 7 -- Statistical MIMO Radar, R. Blum

1600-1630 Talk 8 -- Fast Methods for Real-Time Level-Set Curve Evolution, C. Karl

1630-1700 Talk 9 --Topo-Geometric Modeling: A notion of 2/3D Shapes, H.   Krim

1830-2100 Dinner: University Club, Hillsborough Street, Raleigh, NC

1930-2030 Keynote Address:  Recent Results in Distributed Sensing for Detection and Beamforming, L.  Scharf

27 May 2005

0830-0930 Plenary III   Graphical Models, Distributed Fusion, and Sensor Networks, A. Willsky

0930-1000 Talk 10 -- Methods for simulating rare events in optical systems, Kath

1000-1015 Break: EGRC 246

1015-1100  Talk 11 -- Shape, Shape Matching Metrics, and Learning Shape via Sampling (Shapelets), G. Arnold/Stiller

1100-1130 Talk 12 -- Constrained, Globally Optimal, Multi-frame Motion Estimation, Milanfar

1130-1200 Talk 13 –Information: Theoretic Bounds on ATR Performance from Laser Radar Data, Lanterman

1200 Lunch and adjourn.


Schedule:

Reception: Wednesday May 25th 2005, at EB1 at 6:30PM-8:00PM.
Parking is next door behind Red Hat Building.

Presentations: Thursday-Friday May 26th-27th in EGRC 136

Parking is permitted upon retriving a free parking ticket at the Centennial Campus Entrance booth

Dinner: Dinner with a keynote address will be at the NCSU Faculty Club on Hillsborough St. http://www.ncsuclub.com

Lunch and Breaks: Lunch and Breaks will be held in EGRC 246 at the designated times in the program.

Events:

To be determined.

Abstracts:

Statistical MIMO Radar
Rick S. Blum
Electrical and Computer Engineering Dept.
Lehigh University
Bethlehem , PA

Inspired by recent advances in multiple-input multiple-output (MIMO) communications, we propose the statistical MIMO radar concept. The fundamental difference between statistical MIMO and other radar systems is that classical radar systems, for example, radars employing arrays or space-time adaptive processing (STAP), implement beamforming, which requires a high correlation between signals either transmitted or received by an array, while the proposed MIMO radar exploits the independence between signals at the array elements. It is well known that in conventional radar, target radar cross section

(RCS) fluctuations, particularly slow fluctuations modeled as Swerling Case I, degrade radar performance. The novelty of statistical MIMO radar is that it takes the opposite view, namely, it capitalizes on target RCS scintillations and glint to improve the radar's performance. MIMO radar utilizes multiple antennas at both the transmitter and receiver, and it can be applied in monostatic or bistatic modes. The antennas at each end of the radar system have to be sufficiently separated such that the target provides uncorrelated reflection coefficients between each transmit/receive pair of antennas. Through preliminary analysis and numerical results, we have demonstrated that when the separation of a radar's antennas is such that antennas have uncorrelated views of the target, MIMO radar greatly improves detection and estimation performance due to the absence of target fades. We refer to this gain as diversity gain. In addition to the diversity gain, we have demonstrated that the ability of MIMO to resolve multiple targets spatially leads to improved target resolution in the range-delay domain. This effect is similar to a spatial multiplexing gain.

An Introduction to RHINO (Real-time Histogram Interpretation of Numerical Observations)
Authors: Susan Chandler and Gordon Lukesh
Nukove Scientific Consulting
69 Vista Linda Road
Ranchos de Taos, NM 87557 USA
E-mail: smchandler@nukove.com; gordon.lukesh@nukove.com

RHINO is a software package under development for the United States Air Force Office of Scientific Research. RHINO uses algorithms developed by Nukove to estimate the pointing errors for ground-to-space illumination experiments under the conditions of a far-field beam full-width-half-maximum comparable to the atmospheric and mechanical pointing disturbances common to laser systems. The pointing errors are known as jitter, a shot-to-shot random pointing error, and boresight, a fixed offset due to residual alignment errors. The statistical approach uses predetermined probability distributions for a range of pointing errors expected in ground-to-space illumination experiments. Small data sets are tested via chi square methods against the distributions and the highest statistical confidence Q associated with the chi square tests provides the pointing estimate.

The eventual package will be a real-time system capable of providing immediate feedback to either a laser system operator or an automatic control system to improve system performance, such as reducing boresight error or adjusting the beam size for optimal target illumination. The system will display pointing estimates for an operator and also record aspects of the system performance for post processing.

Multiscale Geometric Signal Processing in High Dimensions
Hyeokho Choi and Richard Baraniuk
Rice University
Houston , Texas

There is an increasing need for efficient representations and processing algorithms for the growing range of applications that involve high-dimensional data. In this talk, we overview two of our recent efforts on extracting, modeling, and processing geometric information hidden in high-dimensional data sets. First, we introduce a new hypercomplex wavelet transform that provides a geometry-friendly and economic representation for higher-dimensional, piecewise smooth signals. Second, we develop a new multiscale representation for image manifolds. Our approach is centered on the observation that, while typical image manifolds are not differentiable, they have an intrinsic multiscale geometric structure.

Free-Space Laser Communications: Propagation and Beam Formation Investigations
M. Giles
New Mexico State University
Las Cruces , NM

The primary objective of this project is to develop new methods to improve the robustness of free-space optical (FSO) communications systems.

During the first year, after considering several ways to achieve this objective, we focused our investigations on two methods. (1) the generation of partially coherent beams that are less susceptible to turbulent, particulate-laden atmosphere, and (2) the use of multiple ground receiver stations at different locations to increase the probability that at least one ground site will always be in communication with an air vehicle (air-to-ground scenario). Investigations at NMSU of method (1) generation of partial coherence, have resulted in the preliminary development of a new static partial coherence generator that utilizes a unique combination of optical fibers to reduce the spatial coherence of a transmitted laser beam. The characteristics of this device will be presented, including laboratory performance data measured at the receiver as the beam propagates through simulated atmospheric turbulence (a variable phase screen). Insertion of the device at the transmitter results in a significant reduction in scintillation detected at the receiver. Nukove Scientific Consulting, a partner with NMSU for this study, is examining the performance improvement of FSO air-to-ground links in the presence of clouds or fog with the use of method. (2) multiple ground receiver stations at different locations. They have constructed a probabilistic model involving varying cloud cover, a number of ground receiver stations with separated locations, and communication with an air vehicle, such as a UAV. The model will be presented.

Optimal probing for system identification with applications to Biomolecular NMR
N. Khanedja
Harvard University
Cambridge , MA

In many military and biological applications of signal processing, one is faced with the situation that the signals recorded by the sensors are largely influenced by how the system of interest is probed or excited. Therefore, besides design of optimal estimators, it is possible to acquire information about the system more efficiently by design of input signals and waveforms that probe the system of interest. These input signals should be so designed that they attempt to reduce the uncertainty
of the unknown parameters the most. In this talk I will present our work on optimal design of radio frequency pulse sequences in high resolution NMR spectroscopy for structural studies of biomolecules. I will derive fundamental limits on the information that can be extracted in these system identification experiments.

Fast Methods for Real-Time Level-Set Curve Evolution
W.C. Karl
Boston University
Boston , MA

Curve evolution methods have become popular in a wide variety of image processing applications. Correspondingly, the level-set method has gained wide acceptance in the numerical implementation of such curve evolution methods. Unfortunately, the relatively high computational cost of level-set methods has been a bottleneck for real-time and near real-time applications. In this talk we present novel and fast implementations of level-set based curve evolution. The methods avoid solving partial differential equations while preserving the advantages of level-set methods, such as automatic handling of topological changes.

In our implementations only simple operations matched to the digital nature of computers and many images are used. In particular, curve evolution is accomplished through fast insertion and deletion operations on two linked grid point lists. Two fast algorithms are presented: one for generic evolution speeds and a second for a special, yet practically important, class of speeds that is composed of a data dependent external speed and an internal, curvature-dependent speed producing regularization through boundary smoothness. Compared with previous optimized narrow-band algorithms, our first algorithm can achieve an order of magnitude speedup, and the second algorithm can achieve two orders of magnitude speedup in both 2D and 3D image segmentation applications. We demonstrate a number of example applications, including a real-time video tracking system which combines our algorithm with the image acquisition toolbox of Matlab to perform faster than real time level-set based object tracking.

Methods for simulating rare events in optical systems
William L. Kath
Northwestern University
Engineering Sciences and Applied Mathematics
2145 Sheridan Road
Evanston , IL 60208-3125
847-491-8784 kath@northwestern.edu

Optical systems transmit and process information at extremely high rates. Errors are handled at slower electronic speeds, however, and thus systems must be designed to have extremely small error rates, typically one error per 10^9 or more bits. Since overall system performance is determined by extremely rare events, the accurate modeling of such systems presents a severe mathematical and computational challenge. The application of importance sampling (a member of a general family of variance reduction techniques) is one method that can overcome this difficulty. In this talk, we will discuss how importance-sampled Monte-Carlo simulations can be used to determine transmission impairments caused by amplified spontaneous emission noise in soliton-based optical communication systems. The method allows numerical simulations to be concentrated on the noise realizations that are most likely to result in transmission errors, leading to speedups of several orders of magnitude over standard Monte Carlo methods.

Information-Theoretic Bounds on ATR Performance from Laser Radar Data
A. Lanterman
Georgia Inst. Of Technology
Atlanta , GA

Laser radar systems offer rich data sets for automatic target recognition. ATR algorithm development for laser radar has focused on achieving real-time performance with current hardware. In comparison, little work has been done on understanding how much information can be squeezed from the data, independent of any particular algorithm. To help fill this gap, will present information-theoretic bounds based on statistical models for laser radar data. For raw imagery, we employ models based on the underlying physics of the detector. For assembled "point clouds," whose statistics are a complex interaction of the underlying sensor characteristics and the jigsaw puzzle algorithms used to assemble multiple views, we consider a Poisson point process model chosen for its analytical tractability.

Most ATR algorithms for laser radar data are designed to be "invariant" with respect to position and orientation. Our information-theoretic bounds illustrate that even algorithms that do not explicitly involve the estimation of such nuisance parameters are still affected by them.

The presentation will also include brief discussions of two other efforts: 1) algorithms for ATR with infrared data, with support code that will be made available to the ATR community, and 2) the assembly of an experimental testbed to explore passive radar systems that exploits "illuminators of opportunity" such as FM radio and television broadcasts.

SHARPENING TECHNIQUES FOR SENSOR FEATURE ENHANCEMENT
Larry Marple
School of EECS
Oregon State University
Eugene , OR

One-dimensional (1-D) techniques based on bandwidth extrapolation (BWE) and bandwidth enhancement (also called BWE) have been developed, beginning in the late 1970s, to add predicted high frequency content to 1-D signals where they were missing these frequencies, and which can be extended to the 2-D case to handle imagery high-frequency restoration. The effect of the techniques is to sharpen the temporal content (1-D signals) or spatial content (2-D images). The past 10 years have seen the development of high-resolution 2-D spectral analysis techniques (for example, linear prediction [LP], autoregressive [AR], minimum variance [MV] methods) and associated fast computational algorithms. These algorithms estimate 2-D LP/AR/MV parameters in the original 2-D signal/image domain, in order to produce high-resolution results in the transform domain (spectral domain).

As a new avenue of research, we would like to switch the domain of application of the 2-D high resolution techniques from the original data domain of application instead to the transform domain, in order to produce sharpened results in the original domain. For example, an image would be 2-D transformed (many choices here: Fourier, wavelet, etc.), a 2-D linear prediction applied to the spatial frequencies, and then a transform-domain spectral analysis performed (effectively taking us back into the original image domain) to produce a sharpened result. Switching the domain of application will require some adjustments to the current state-of-the-art in 2-D high resolution spectral analysis techniques. This paper will review this new research approach, with particular attention to sensor processing in 2-D and 3-D applications.

Constrained, Globally Optimal, Multi-frame Motion Estimation
P. Milanfar
University of California
Santa Cruz , CA

In address the problem of estimating the relative motion between the frames of a video sequence. In comparison with the commonly applied pairwise image registration methods, the proposed method considers global consistency conditions for the overall multi-frame motion estimation problem, and is more accurate. We review the recent work on this subject and propose an optimal framework, which directly applies the consistency conditions as both hard constraints in the estimation problem, or as soft constraints in the form of stochastic (Bayesian) priors. The framework is applicable to virtually any motion model and enables us to develop a robust approach, which is resilient against the effects of outliers and noise.

SENSING AND COMMUNICATIONS USING ULTRAWIDEBAND RANDOM NOISE WAVEFORMS
Professor Ram M. Narayanan
Department of Electrical Engineering
The Pennsylvania State University
University Park , PA 16802
Tel: 814-863-2602
Email: ram@ee.psu.edu

Deterministic signals such as pulsed and frequency-modulated waveforms have traditionally been used for sensing and communications. While such waveforms are mathematically tractable for modeling and analysis, they suffer from poor electronic counter measures (ECM) properties in the battlefield. Recently, we have developed and refined the use of ultrawideband (UWB) random noise waveforms for sensing and communications applications. The use of UWB signals provides high range resolution for sensing and spread spectrum capability for communications. These signals have excellent ECM properties, such as low probability of intercept (LPI), low probability of detection (LPD), and immunity to jamming and interference. In addition, noise radar and noise communications systems are low cost, lend themselves to easy hardware implementation, and require simplified signal processing at the RF stage. The narrowband representation for a zero-mean wide-sense stationary random process is used to model the UWB noise signal. This allows us to easily follow the RF signals through various components and to understand the signal characteristics at different locations within the system. The receiver signal processor uses the technique of heterodyne correlation wherein a cross-correlation is performed between the received waveform and (i) a frequency-shifted and time-delayed transmit waveform replica for sensing applications, and (ii) a frequency-shifted and diversity-enhanced transmit waveform replica for communications applications. The heterodyne correlation receiver functions as a matched filter and permits the eventual down conversion of the UWB received signal to a narrow frequency band, thereby enhancing the overall signal-to-noise ratio (SNR) and retaining phase coherence. Demonstrated applications include polarimetry, interferometry, SAR/ISAR imaging, foliage and wall penetration imaging, and covert communications.

A Review of Recent Results in Distributed Sensing for Detection and Beamforming
Louis Scharf, Edwin K.P. Chong, John A. Gubner*, Ali Pezeshki
Colorado State University, Fort Collins, CO
* University of Wisconsin, Madison, WI

Our aim will be to consider three general classes of problems: (1) likelihood fusing of detection statistics from distributed sensors, and connections with ACE (adaptive coherence estimator) and capacity, (2) one-bit fusing of likelihood and the dependence of its error probability on SNR and bandwidth, and (3) recent results on matched and adaptive subspace detectors for beam forming of imperfectly modeled fields.

 

Housing:

To make a reservation, one should mention AFOSR REVIEW MEETING  May 25-27, 2005 to get the special room rate of $75.00.

http://www.ichotelsgroup.com/h/d/hi/1/en/direction/rduhs 

TO MAKE RESERVATIONS:  reser@brownstonehotel.com

Holiday Inn - Brownstone
1707 Hillsborough Street
Raleigh, NC 27605
(919) 828-0811
1-800-331-7919
1-800-465-4329

Directions:

From Airport (Raleigh-Durham International) to Brownstone-Holiday Inn Hotel
http://www.ichotelsgroup.com/h/d/hi/1/en/direction/rduhs

From Brownstone-Holiday Inn Hotel to Centennial Campus: Directions.pdf

How to get to NC State university Centennial Campus
Map of Centennial Campus:
http://centennial.ncsu.edu/contact/zoommap2.pdf
(please note that the Monteith building is still refered to as EGRC on this map)

Registration:

Pre-registration is required.

Please Complete and Return Form: AFRegistration.pdf
By email to Elaine Hardin at elainehardin@ncsu.edu
Or by fax to Elaine Hardin at 919-515-5523

DEADLINE: May 20, 2005

For more information about this workshop or to be included in our E-mail list, please contact Hamid Krim at ahk@ncsu.edu