Increasingly, for many application areas, it is becoming important to include elements of nonlinearity and non-Gaussianity in order to model accurately the underlying dynamics of a physical system. Moreover, it is typically crucial to process data on-line as it arrives, both from the point of view of storage costs as well as for rapid adaptation to changing signal characteristics. In this paper, we review both optimal and suboptimal Bayesian algorithms for nonlinear/non-Gaussian tracking problems, with a focus on particle filters. Particle filters are sequential Monte Carlo methods based on point mass (or "particle") representations of probability densities, which can be applied to any state-space model and which generalize the traditional Kalman filtering methods. Several variants of the particle filter such as SIR, ASIR, and RPF are introduced within a generic framework of the sequential importance sampling (SIS) algorithm. These are discussed and compared with the standard EKF through an illustrative example.

/pdf/a-tutorial-on-particle-filters-for-online-nonlinear-non-1rzkwgk2pp.pdf

A tutorial on particle filters for online nonlinear/non-Gaussian Bayesian tracking

Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

Today's Web-enabled deluge of electronic data calls for automated methods of data analysis. Machine learning provides these, developing methods that can automatically detect patterns in data and then use the uncovered patterns to predict future data. This textbook offers a comprehensive and self-contained introduction to the field of machine learning, based on a unified, probabilistic approach. The coverage combines breadth and depth, offering necessary background material on such topics as probability, optimization, and linear algebra as well as discussion of recent developments in the field, including conditional random fields, L1 regularization, and deep learning. The book is written in an informal, accessible style, complete with pseudo-code for the most important algorithms. All topics are copiously illustrated with color images and worked examples drawn from such application domains as biology, text processing, computer vision, and robotics. Rather than providing a cookbook of different heuristic methods, the book stresses a principled model-based approach, often using the language of graphical models to specify models in a concise and intuitive way. Almost all the models described have been implemented in a MATLAB software package--PMTK (probabilistic modeling toolkit)--that is freely available online. The book is suitable for upper-level undergraduates with an introductory-level college math background and beginning graduate students.

Machine Learning : A Probabilistic Perspective

Monte Carlo methods are revolutionizing the on-line analysis of data in fields as diverse as financial modeling, target tracking and computer vision. These methods, appearing under the names of bootstrap filters, condensation, optimal Monte Carlo filters, particle filters and survival of the fittest, have made it possible to solve numerically many complex, non-standard problems that were previously intractable. This book presents the first comprehensive treatment of these techniques, including convergence results and applications to tracking, guidance, automated target recognition, aircraft navigation, robot navigation, econometrics, financial modeling, neural networks, optimal control, optimal filtering, communications, reinforcement learning, signal enhancement, model averaging and selection, computer vision, semiconductor design, population biology, dynamic Bayesian networks, and time series analysis. This will be of great value to students, researchers and practitioners, who have some basic knowledge of probability. Arnaud Doucet received the Ph. D. degree from the University of Paris-XI Orsay in 1997. From 1998 to 2000, he conducted research at the Signal Processing Group of Cambridge University, UK. He is currently an assistant professor at the Department of Electrical Engineering of Melbourne University, Australia. His research interests include Bayesian statistics, dynamic models and Monte Carlo methods. Nando de Freitas obtained a Ph.D. degree in information engineering from Cambridge University in 1999. He is presently a research associate with the artificial intelligence group of the University of California at Berkeley. His main research interests are in Bayesian statistics and the application of on-line and batch Monte Carlo methods to machine learning. Neil Gordon obtained a Ph.D. in Statistics from Imperial College, University of London in 1993. He is with the Pattern and Information Processing group at the Defence Evaluation and Research Agency in the United Kingdom. His research interests are in time series, statistical data analysis, and pattern recognition with a particular emphasis on target tracking and missile guidance.

/pdf/sequential-monte-carlo-methods-in-practice-3kfxeebn8e.pdf

Sequential Monte Carlo methods in practice

Planning and navigation algorithms exploit statistics gleaned from uncertain, imperfect real-world environments to guide robots toward their goals and around obstacles.

Probabilistic Robotics

We describe an adaptive coded-aperture imager operating in the midwave IR. This consists of a coded-aperture mask, a set of optics, and a 4k×4k focal plane array (FPA). This system can produce images with a resolution better than the detector pixel limit by combining multiple frames of data recorded with different coding. This superresolution capability has been demonstrated both in the laboratory and with targets placed outside, the highest resolution being one-half of the FPA pixel pitch.

Adaptive coded-aperture imaging with subpixel superresolution.

The use of silicon substrates has been very successful for producing large area focal plane arrays operating in the MWIR waveband using the MBE growth process. More recently, promising results have been obtained in the LWIR waveband using a MOVPE growth process on a buffered silicon substrate. The MOVPE growth process is also suitable for more complex multi-layer structures and we have now used this technique to produce our first MW/LW dual waveband focal plane arrays. In this paper we show that close to background limited performance can be achieved in both wavebands, however the main challenge with arrays grown on silicon is to obtain low defect counts. These first arrays are promising in this respect and operabilities of 99.4% and 98.2% have been achieved in the MWIR band and LWIR band respectively. The availability of dual waveband arrays allows the correlation of defects in the two wavebands to be compared. In general, we find that the correlation is low and this suggests that defect generation mechanisms which would affect both bands (such as threading dislocations) are currently not the main source of defective devices in MOVPE grown devices on silicon.

http://client.blueskybroadcast.com/spie/dss08/pdf/6940-88-a5b_hall_paper.pdf

Dual waveband MW/LW focal plane arrays grown by MOVPE on silicon substrates

Optical coherence tomography (OCT) is increasingly used in areas such as ophthalmology and contact lens metrology. However, in such cases, image distortion can occur due to the non-planar nature of the measured sample. Postprocessing algorithms can be implemented to correct this distortion. Here we present an OCT phantom designed to confirm the validity of post-processing algorithms used for measuring curved surfaces. A multi-purpose OCT phantom has been created within a fused silica plano-convex lens using the direct femtosecond laser writing technique. This phantom can be used to calibrate and quantitatively assess the performance (e.g. resolution, sensitivity and distortion) of OCT systems and associated post-processing algorithms for curved structures such as lenses. This novel OCT phantom has been characterized using an optical microscope and OCT systems.

/pdf/non-planar-calibration-phantoms-for-optical-coherence-gk8fbp8umt.pdf

Non-planar calibration phantoms for optical coherence tomography

Adaptive coded aperture imaging (ACAI) has the potential to enhance greatly the performance of sensing systems by
allowing sub detector pixel image and tracking resolution. A small experimental system has been set up to allow the
practical demonstration of these benefits in the mid infrared, as well as investigating the calibration and stability of the
system. The system can also be used to test modeling of similar ACAI systems in the infrared. The demonstrator can use
either a set of fixed masks or a novel MOEMS adaptive transmissive spatial light modulator. This paper discusses the
design and testing of the system including the development of novel decoding algorithms and some initial imaging
results are presented.

An experimental infrared sensor using adaptive coded apertures for enhanced resolution

We have previously discussed the potential of using a Hg1-xCdxTe source as a reference plane for the non-uniformity correction of thermal imagers and which is being developed as an option for the UK 3 rd generation, high performance thermal imaging program (Albion). In this paper we will present our first results on a large area (1.5 cm x 1.5 cm) source which was grown on a silicon substrate and can simulate a range of temperatures from -10 o C to +30 o C. Due to the fast switching speed, the apparent temperature can be changed on a frame by frame basis. Also, the operation of the device can be synchronized to the integration time of the camera to reduce the mean power requirements by a factor of 10 and reduce thermal heating effects. The main applications for Hg1-xCdxTe devices as high-performance, cryogenically-cooled detectors typically require very low drive currents. The use of this material for large-area LEDs has generated new challenges to deal with the high peak currents. These are typically in the range 1-2 A/cm 2 for a MWIR waveband source and have led to a need to reduce the common impedance, reduce the contact resistances and consider the effects of current crowding.

Neil Gordon

Papers

Adaptive coded-aperture imaging with subpixel superresolution.

Dual waveband MW/LW focal plane arrays grown by MOVPE on silicon substrates

Non-planar calibration phantoms for optical coherence tomography

An experimental infrared sensor using adaptive coded apertures for enhanced resolution

A high-speed, MWIR reference source for FPA non-uniformity correction using negative luminescence