Robust Control Theory and Design
Robust Control Applications
Predictive Control and Identification
Signal Processing and Communications
Speech Processing and Neural Networks
Medical Applications of Information Engineering
Computer Vision and Robotics
Artificial Intelligence and Decision Support
Departmental Computing Systems
Prof. K. Glover
Dr M.C. Smith
Dr G. Vinnicombe
Work has continued on theoretical enhancements to the H-infinity; robust control framework that are aimed at making the theory more useful in a design context.
A substantial new graduate level text book(I130) has been completed and covers many of the developments in robust control developed over the last decade in Cambridge, by the co-authors from Caltech and from elsewhere.
Further results on the robust control of sampled data systems have been presented(I20,I21,I66).
Research effort has continued on input-output approaches to control systems(I107). A generalisation of the gap metric for uncertainty studies in nonlinear feedback systems has been developed(I30,I31). Approaches to the computation of induced norms for nonlinear systems were pursued(I98,I99). The optimality properties of H-infinity; loop-shaping designs were characterised in the language of classical frequency response control(I22).
The results of H-infinity; robust control can often appear conservative when viewed from a practical perspective. Techniques for reducing this apparent conservatism, which exploit bounds on the controller complexity, are developed in(I119). A generally applicable method for dealing with actuator saturation within a robust control framework is presented in(I87). H-infinity; optimization over fixed structure controllers remains a difficult problem - a useful reparameterization of the problem, with some encouraging results, is reported(I86).
Prof. K. Glover
Work on the control of combustion engines has expanded substantially in the last year with a new experimental facility under development in the department with EPSRC and Ford Motor Company funding. This is a continuation of current work on the robust control of air-fuel ratio so as to minimize emissions.
Work on flight control has seen a significant input into the GARTEUR design challenge from the group using the H-infinity; Loop Shaping design procedure.
Dr J. Maciejowski
Research on Model-based predictive control (MBPC), which allows economic and engineering constraints, as well as equipment failures, to be taken into account when controlling complex plant, has continued(I41). The applicability of this approach to flight control is being investigated by continued participation in the GARTEUR `Flight Control Challenge'.
Work on the application of balanced realisations to system identification has continued within the framework of the EC-sponsored `European Research Network into System Identification', which continues to exist as an informal grouping of researchers, although EC funding has now terminated(I77). We have also investigated ways of ensuring the stability of models obtained with the so-called `subspace methods', which are becoming extremely popular for identifying multivariable systems(I13,I76).
Work has continued on improving facilities for computer-aided control engineering(I100), in collaboration with the Institute of Mining and Metallurgy, Krakow, Poland.
Dr R.J. Richards
A new approach to image processing, partial summation, has been developed, analysed and applied to a variety of real problems. Partial summation allows many spatial image processing operations to be achieved very efficiently on a conventional processor, which is of particular benefit in real-time industrial vision applications(I43). A novel, very low cost robotic system has been developed and applied to the task of automated PCB assembly. The inaccuracies of a low specification mechanical system can be overcome with the use of a sophisticated controller, which includes visual feedback from the robot end-effector. Partial summation allows the necessary image processing to be implemented in real-time on a standard PC. Work has progressed on developing a theoretical framework to assess and accelerate reinforcement learning techniques, with particular reference to robot and flexible part movement. This system has been implemented for the time-optimal robotic transportation of flexible materials.
Dr W.J. Fitzgerald
Dr S.J. Godsill
Dr N.G. Kingsbury
Dr M.D. Macleod
Dr S.J. Maric
During the year work has continued with mud pulse telemetry and better coding and demodulation schemes have been developed. These methods have been tested on real data in collaboration with Schlumberger Cambridge Research Ltd.
As part of the continuing research into the restoration of degraded manuscripts using image processing and model based methods(I48,I110), some of the models (HMM's and the Viterbi algorithm) have been applied, with excellent results, to the problem of sequence alignment and dendrochronology. This work will continue and will be applied to other areas were sequence alignment is a problem, e.g. DNA sequencing. This year a proposal for a six month programme at the Isaac Newton Institute has been accepted and it will take place in 1998. The theme of the programme is The Mathematics of Nonlinear and Nonstationary Signal Processing.
Work has continued with Bayesian methods applied to a variety of problems in time series analysis(I23,I89,I92,I93). In particular, the problem of Non-Gaussian statistics has been addressed and the family of alpha-stable processes has been investigated(I64,I65).
The analysis of sea clutter data in collaboration with the DRA has produced very good results.
Several international collaborations have started this year (in addition to our existing ones) with the Institute of Statistics and Decision Sciences at Duke University, INRIA in France and the Numerical Mathematics Department at the University in Vienna. These collaborations address Sequential MCMC methods, SAR images and Random Markov fields and Time-Frequency methods (Gabor frames and wavelets) respectively.
Work on biological signal processing has continued on ion channel dynamics and change point detection in collaboration with the Pharmacology department(I42), the analysis of the firing of neurons in the brain and their relationship with certain muscular activities in collaboration with the Institute of Neurology, and the analysis of optical microscope images for the detection of abnormal cells in collaboration with a local company.
Research has continued in the area of signal analysis and enhancement in the presence of noise. In the processing of audio signals work has focused on model-based statistical methods for treatment of signals degraded by non-Gaussian noise using both the Expectation-Maximize (EM) algorithm(I35) and Markov Chain Monte Carlo (MCMC) methodology(I32,I33,I35,I36,I37). This section of work saw the completion of a long and fruitful involvement with the British Library, which had funded the work in sound and image restoration for a number of years.
In the signal separation area progress has been made in separation of acoustical signals from convolutive mixing environments using decorrelation techniques(I10,I11,I12), achieving some excellent audio results. As a consequence of the success of this work Olivetti Research Limited have proposed funding a research project in separation techniques for the multimedia office environment.
The work in Motion Picture Restoration(I61,I62,I90) has evolved to produce a number of new algorithms for the reconstruction of entire missing frames. Further improvements in detail preservation have been attained through the use of a Bayesian strategy for video sequence modelling. Results employing new fast relaxation techniques have been reported(I34,I60,I63).
The final results from a project on the design of multidimensional wavelet filter banks were published(I115), with emphasis in this case on the application of transformation-of-variables methods to IIR digital filters.
Projects on methods of modifying standard image and video compression methods, so that they are suitable for use over noisy and error-prone transmission channels, have been progressing well. The basic error resilient entropy code (EREC) algorithm has been published in detail(I94), and the algorithm has been successfully applied to MPEG-II standard video bit-streams(I111,I112), to enhance the resilience to transmission errors while requiring no increase in transmitted data rate or bandwidth.
Motion estimation for video signals has been a key feature of several projects. In particular, wavelet techniques have been extended to use filter banks with complex coefficients so that object motion can be measured to high accuracy within a multi-resolution framework(I57,I79,I80). Enhancements to these methods have been developed(I78) to provide greater accuracy with reduced computation.
Other image processing projects in progress are concerned with: acquisition of 3-D data of human faces to help in the task of face recognition; efficient coding of data from multiresolution image transforms using lattice vector quantisers; and improved motion compensation and segmentation methods for object-based video compression.
During this year considerable progress has been made in developing a framework in which 3D projective invariants and constraints relating points in multiple views of a scene can be formulated(I2,I68,I69). We now intend to use these invariants and tensorial relations for matching features over a number of images. More work has also been done on the estimation of the structure of a scene and the motion of the camera from two images of the same object taken with a moving camera(I70).
Work on the use of linear functions as an alternative to matrices and tensors has continued. The methods used allow us to encode structure into the linear functions and thus impose this structure during optimization processes. Some applications of these techniques include the estimation of power spectra and the separation of audio signals.
Work on general applications of geometric algebra to engineering problems is continuing(I67).
The research in general area of wireless communications has been divided into two parts. In the theoretical part we have looked into the basic properties of frequency hopping CDMA signals(I81,I83) which are used in wireless networks. The goal of this work is to provide better understanding of the theory of spread spectrum as a technique applied in wireless communications. For this we have used the fundamental definitions from information theory like the Shannon bandwidth and channel capacity. Research in practical aspects of wireless communications has centered around new ways of increasing system's capacity, as well as looking forward into new generation of wireless systems which will be able to provide multimedia services(I84). The work in wireless communications was closely related to our research in optical communications(I82,I85) since we believe that the optical networks will be used as a backbone network for wireless communication systems.
Further work has been published on reduced complexity multiplier design(I18) and digital filter design(I19). New results have been published on interpolators for variable delay implementation(I72). Research into high resolution estimation continues(I129). Research has started into the use of "turbo-codes" in radio communications.
Prof. S.J. Young
Dr A.H. Gee
Dr M. Niranjan
Dr R.W. Prager
Dr A.J. Robinson
Mr P.C. Woodland
The work in speech over the last year continued to focus on large vocabulary speech recognition based on the HTK HMM system(I116,I126) and the ABBOT connectionist system(I95).
In November 1995, the Speech Group participated once again in an international evaluation of large vocabulary speech recognition systems organised by the US Advanced Projects Research Agency (ARPA). This was the third successive year that CUED has taken part and this time the focus was on transcribing read speech input via unknown microphones in noisy conditions. As in the previous two years, the top-ranked system came from the Cambridge group(I125).
A major contribution to the success of the ARPA evaluation system was the continuing development of techniques for adaptation of a speech recognition system to the current speaker and acoustic environment(I24,I124). The techniques developed in this area have rapidly been taken up by many other groups worldwide.
The final stage of our work on "Improving the Phonetic Discrimination of HMM-based Speech Recognisers" funded by the EPSRC was completed with a study of methods for using Maximum Mutual Information-based training in large vocabulary systems(I117,I118).
Projects with industrial collaboration were also completed last year. The DTI/EPSRC/Olivetti "Video Mail Retrieval using Voice" project ended successfully with a demonstration retrieval system which is both speaker and topic independent(I49,I50,I51,I52,I53). Although focussed originally on the retrieval of video mail messages, this work is now being extended to the broader task of retrieving broadcast news programmes(I4). A similar project sponsored by Hewlett-Packard was also completed with the final phase focusing on fast implementation methods for word spotting(I58,I59).
PhD project work continues with good progress being made in the areas of speech recognition in noise(I25,I26) and stochastic speech production modelling(I3) and class-based language models(I91).
Research in connectionist speech recognition continues with theoretical advances relating to the training of a recurrent network within a hidden Markov model(I106) and incorporation of phonetic context within this framework(I55). Practical application was made to the recognition of speech from unknown microphones(I56) and to real-time recognition of broadcast radio speech(I116). An overview of the complete system appeared as a book chapter(I95).
A statistical model called a "Hierarchical Mixtures of Experts" was further developed in terms of parameter estimation algorithms(I120,I121).
Work has continued in vocal tract modelling and accurate parameter estimation in nonstationary environments(I5,I101).
Work on sequential parameter estimation in neural networks has continued. A three year EPSRC funded project `Incremental Learning and Adaptive Neural Network Architectures' was completed during this period(I44,I45,I88,I89). Several applications of the algorithms developed in this area are being actively pursued in areas including enhancing noisy speech and modelling financial time series.
The work on the use of neural networks to solve combinatorial optimization problems culminated in an invited tutorial at the 1995 International Conference on Neural Networks(I27). Recurrent neural networks have been designed to implement a wide variety of low entropy coding schemes(I39,I40). The group also organised the Cambridge Neural Networks Summer School 1996. This was a great success, with seventy participants drawn from industry and academia. The EPSRC again provided thirty scholarships to enable research students from other UK universities to attend. As in the past, the speakers included the foremost experts in the field drawn from all over the world.
Dr A.H. Gee
Dr M. Niranjan
Dr R.W. Prager
Work continues on the EuroPunch and QAMC projects funded by the European Community. These projects are aimed at using artificial neural networks and other statistical analysis techniques to predict risk during pregnancy(I73,I74,I75). The Department's main partner in these projects is the University of Cambridge Department of Obstetrics and Gynaecology.
The EPSRC funded Stradivarius project concluded successfully. This work, in collaboration with the University of Cambridge Department of Radiology, involved combining information from a number of two-dimensional ultrasound scans to produce three-dimensional displays of internal organs and structures(I96,I97,I113,I114). SOLUS-3D, a major follow-up project funded by the EC, commenced in May 1996. This project aims to extend the state-of-the-art in 3D ultrasound and identify practical applications in the areas of obstetrics and gynaecology. SOLUS-3D is a collaborative venture between the Cambridge Departments of Engineering, Radiology and Obstetrics & Gynaecology, as well as other medical and engineering sites across Europe.
The study of the response of liver transplant patients to immunosuppressive drugs continues through the project, "Neural Computing for Nonstationary Medical Signal Processing". It is funded by the EPSRC and involves collaboration with the University of Cambridge Department of Clinical Biochemistry and the Transplant Unit at Papworth Hospital.
Dr R. Cipolla
Dr A.H. Gee
Dr N. Pears
Research in computer vision and robotics has continued on the development of new theories for recovering the three-dimensional shape of visible surfaces from images taken from arbitrary viewpoints with uncalibrated cameras.
Novel contributions have been made in the analysis of curved surfaces (where the dominant image feature is the silhouette or profile). It is well-known that it is possible to reconstruct the shape of curved surfaces from the family of outlines obtained by looking at them from different but known viewpoints. We have also been able to show that viewer motion can be recovered from the envelope of consecutive contour generators - the fronter(I1). These theories have been successfully implemented, thus allowing the registration of images taken from different viewpoints and the reconstruction of arbitrarily complex shapes. The implementation required the detection and tracking of image contours(I9). Computationally simple methods using B-splines were developed to accurately extract image curves and track their temporal evolution .
Progress continues in developing new techniques to match image curves over arbitrary viewpoints for extracting symmetry axes and recognition. Novel geometric invariants under affine and projective transformation groups were developed which to do not suffer from the problems of noise sensitivity of differential invariants and the occlusion problems of global moment invariants(I103). A quasi-invariant parameterisation of image curves which approximates group invariant arc-length with lower spatial derivatives has been shown to be extremely powerful in detecting symmetry in natural and textured images and efficiently recognising arbitrarily complex curves under arbitrary viewpoints and occlusion(I102).
Visual motion, as perceived by a camera mounted on a robot moving relative to a scene, can be used to compute the robot motion and the structure of the scene and aid navigation(I71). Simple qualitative cues such as time to contact and relative surface orientation could be reliably extracted from the image divergence estimated from the temporal evolution of the apparent area of a closed contour(I17,I105,I109). The method was extended to image texture, abundant in natural images, and used to guide a robot manipulator(I104).
Stereo vision is often used to recover 3D position but requires correspondence of image features and calibration of position of cameras. Our research has aimed to find simpler, calibration-free cues to surface position and geometry. A simple but more robust approximation to stereo has been proposed exploiting cues present in orthographic projection only(I38). A hand-eye coordination system has been developed to guide a robot manipulator to pick up unfamiliar objects in unstructured environments with uncalibrated stereo cameras. By detecting and tracking a human hand it is possible to allow the user to point at an object of interest and guide a robotic manipulator to pick it up(I14).
Work continues on the design of novel man-machine interfaces. Pointing and face gestures can provide more natural ways of communicating with computers and machines. Automatic techniques for localising faces(I127,I128) and a realtime visual tracking algorithm(I28,I29) are being exploited in hands-off computer and robot interfaces for physically handicapped people(I15) as well as for an interface to 3D television displays.
The Fourth European Conference on Computer Vision(I6,I7) and an International Workshop on Computer Vision in Human-Machine Interfaces were held in Cambridge in April 1996 and attracted over 360 delegates. An international prize was received for research on the perceptual grouping of image curves(I8).
Dr T. Holden
Dr P. Wilhelmij
Work is progressing in applying Artificial Intelligence and Decision Support techniques to large-scale industrial systems, in particular to energy and safety-critical situations. This work involves wide-ranging participation with international industry and has been supported by donations of equipment from BP and Shell.
Work with BP and Honeywell has led to the development of information and management structures for large, complex information systems for improving performance and reducing risk through better informed operators. Through the LIFETRACK project(I47), a decision-support system has been installed in a BP operational plant. This work follows formats and practices required by CALS (Continuous Acquisition and Lifecycle Support) standards.
Related research work has concerned knowledge-sharing in a corporate environment. A key concern in safety-critical or otherwise risk-prevalent organisations is the management of the human, communication, informational and knowledge factors that influence the quality of decisions(I46,I122). One aspect of this has been the development of methods for supporting technology and skills transfer between a fully developed plant in the UK and a newly-installed one in Malaysia(I123). Particular insight was gained into how to transfer knowledge across cultures.
Mr J.M.R. Matheson
The computing system for teaching remains largely unchanged in structure with only minor upgrades to support larger engineering packages, increased filespace requirements and rolling replacement of the older equipment. Support for Windows as well as Unix based applications is being introduced to broaden the usefulness of the system.
The main growth continues to be in research computing and a Computer Officer has recently been appointed to provide more formal support for this. Administrative computing is also of increasing importance to our efficient operation and for information transfer; considerable effort is being put into supporting these developments.
It is likely that, starting this year, there will be a substantial improvements to the Department's networking infrastructure to support the greatly increasing data transfer needs in all areas.
Information Engineering References