No description available.

Driving at night is a difficult task. In an attempt to ease this task, most automotive companies are developing systems that aim to increase the safety of the driver and his/her passengers at night. Jaguar Cars Ltd have been involved in such project for several years and have developed a Night Vision System (NVS) based upon the Near Infrared (NIR) and Head-Up Display (HUD) technologies. This thesis is concerned with the application of digital image enhancement algorithms to further increase the driver's visual range at night. The purpose of this research work is to provide the driver with a safe and non-disturbing, enhanced view of the road scene ahead, which is presented on a head-up display. In this automotive environment, specific requirements such as real-time processing, robustness and reliability must be kept in mind to design algorithms that will not compromise the safety of the driver, his/her passengers and other road users. To fulfill these requirements, we have developed a novel intelligent image enhancement scheme for night time driving that actively adapts to the road scene. This scheme results in the enhancement of the contrast in a portion of the projected HUD road scene as if extra headlamps were directed to the region of the image that represents where the road is going. Human Factors studies have shown that this region is where the driver is concentrating his attention when driving. The position of the region of interest is defined by the computation of an approximation of the vanishing point of the road, updated for every frame using a novel, reliable and optimised road edge detection algorithm. The enhancement of the contrast within the region of interest is obtained by applying several novel low-level algorithms based upon the grey level segmentation of the image into regions and the use of the global histogram equalisation and quantised bi-histogram equalisation algorithms. These novel algorithms have all been implemented on the Matrox Genesis board based upon the multitasking, multiprocessor and parallel DSP TMS320C80 chip from Texas Instruments. All algorithms described in this thesis are able to sustain real-time processing at the NTSC frame rate of 30 frames per second. This new concept for a night time driving aid is an attractive solution that meets the numerous requirements driven by Human Factors research in an automotive environment, in particular safety requirements.

No description available.

Previous studies of the induced current impedance imaging technique have been unable to reconstruct images of three dimensional (3-D) structures. In this study the cause of the problem is identified and the reconstruction algorithm of Purvis is adapted to facilitate the correct reconstruction of images of a limited class of structures which have the form of a long cylinder. The images produced by the algorithm are improved by a data filter based on that of Barber, Brown and Avis. By consideration of the underlying field equations which govern 3-D induced current Electrical Impedance Tomography (EIT) systems, the finite element method (FEM) is used for the computation of the potential field for arbitrary conductivity distributions excited by various coil configurations. A phantom system is built to test the results of the FEM and particular attention is paid to the improvement of the instrumentation. A statistical comparison of the results of measurement and simulation is unable to detect any error in the FEM model. The FEM model is consequently used to develop the reconstruction algorithm but physical measurements are also used to test the algorithm in the presence of noise. The behaviour of the 3-D algorithm is tested for its plane selectivity showing Similar characteristics to those of injected current systems developed by other workers. A possible approach which could both reduce the volume to which the system is sensitive and generate extra measurements for the possible reconstruction of multi-layered images is investigated.

The computational demands of real-time image processing often dictate the use of techniques such as parallel processing to meet required performance. This thesis considers a range of technology which may be used to accelerate image processing operations. An occam compiler is ported to a PowerPC based parallel computer. A multiprocessor configuration tool and Run Time System is developed, allowing occam programs to be distributed over an arbitrary sized network of PowerPC microprocessors. Code optimization techniques for image processing operations are investigated, with the development of a post-compilation code optimizer. The optimizer provides performance increases between 37% and 450% for a variety of image processing algorithms. The applicability of these tools is demonstrated with two image processing applications, micro-biological rapid imaging and sediment texture analysis. Edge detection, region merging and shape analysis algorithms are discussed in the context of the applications. The image processing algorithms are implemented in occam and performance is compared on serial and parallel platforms. The algorithms are then ported to a hardware implementation in a custom computing device, based on a field programmable gate array (FPGA), using the Handel hardware compilation system. The issues involved with this porting are discussed, including the compromises which must be considered when designing for a size constrained hardware platform. Amongst the issues considered are restricted precision data, low level parallelism and algorithmic simplifications. To provide performance equivalent to the hardware, between 5 and 10 processors would be required on the parallel machine, with considerably greater cost, size and power consumption.

The development of a surface texture evaluation technique for the study of roughnesses of the order of a few microns using the signal processing of ultrasonic pulse-echo signals is described. The technique of extracting surface information by means of deconvolution is introduced. Strictly, a solution to the deconvolution problem normally does not exist or is not unique. The chosen method of approaching a solution is by the nonlinear Maximum Entropy Method (MEM), which offers superior image quality over many other filters. The algorithm is described and translated into a standalone computer programme-the development of this software is described in detail. The performance of the algorithm in the field of ultrasonics is assessed by means of the study of simulations involving images similar to those obtainable in a real application. Comparison with the linear Wiener-Hopf filter is provided particularly in instances where the comparison shows weaknesses of either technique. Also examined is the frequency restoration property of the algorithm (not shown by the Wiener-Hopf filter)-potential applications of this property are also described. The final part of the study of the MEM is an examination of the effect on performance of some of the algorithm's parameters and on computer system dependencies. A brief overview of some of the surface metrology techniques currently used is given. The aim is an introduction to surface metrology and an assessment of where the technique described here fits into the general surface metrology field. The experimental system, which of course is essential to practical applications, is considered in some detail. Also considered is a wide range of ultrasonic transducers available for the research. These show a considerable variety of characteristics. Some assessment is carried out using the Maximum Entropy Method with simulated and real data to try and establish the properties of a transducer best suited to the application intended. Finally, results from grating-type test surfaces and more general rough surfaces are presented. The former are intended as a means of establishing the potential performance of the technique; the latter build on the grating results to analyse real surfaces as made by a variety of engineering techniques. Results are compared with those obtained by a stylus instrument. Generally good agreement is found, with roughnesses of around 2 microns being accurately assessed. With the accuracy of these results being less than a micron, it is concluded that this technique has a valuable contribution to the surface metrology field.

No description available.

No description available.

No description available.

This thesis presents an Artificial Neural Networks (ANNs) based, automatic Pattern Recognition(PR) technique for electrical machines Condition Monitoring(CM) applications. The performance of synchronous generators has been studied under a variety of conditions and monitored using a range automatic pattern recognition approaches. The harmonic components of the generator stator, rotor and excitation currents have been analysed initially to gain information of fault conditions in the machines, and then as a source of data for input training patterns to the neural nets. Artificial neural networks; their architecture, algorithms and their application to pattern recognition have been studied. Two unsupervised self-organising neural networks were chosen for further investigation and applied to the automatic pattern recognition tasks. These two neural network models can be classified as Kohonen neural nets and Adaptive Resonance Theory nets. A computer implementation of Kohonen Self Organising Feature Maps(KSOFM) and a simulation that interprets the continuous valued model of adaptive resonance theory (ART2 net), have been studied in detail. General condition monitoring techniques for electrical machines have been briefly reviewed and statistical pattern recognition methods have also been described. To confirm the utility of the proposed ANNs based automatic pattern recognition techniques for electrical machine condition monitoring, two synchronous generators with different capacities, one of 8kva was used for training the networks, and another of 11kva for testing the networks, were employed in the experimental study. The stator, rotor and excitation current signals of a generator have been used to provide the networks' input patterns, and Kohenon networks and adaptive resonance theory networks applicability to electrical machines condition monitoring compared. The possibility of using the proposed techniques to real industrial systems has been discussed. Finally, some of the difficulties of implementing ANNs for condition monitoring are considered.