Parallel Computer Vision Algorithms for Graphics Processing Units

La evolución de los teléfonos móviles inteligentes, dotados de cámaras digitales, está provocando una creciente demanda de aplicaciones cada vez más complejas que necesitan algoritmos de visión artificial en tiempo real; puesto que el tamaño de las señales de vídeo no hace sino aumentar y en cambio el rendimiento de los procesadores de un solo núcleo se ha estancado, los nuevos algoritmos que se diseñen para visión artificial han de ser paralelos para poder ejecutarse en múltiples procesadores y ser computacionalmente escalables. Una de las clases de procesadores más interesantes en la actualidad se encuentra en las tarjetas gráficas (GPU), que son dispositivos que ofrecen un alto grado de paralelismo, un excelente rendimiento numérico y una creciente versatilidad, lo que los hace interesantes para llevar a cabo computación científica. En esta tesis se exploran dos aplicaciones de visión artificial que revisten una gran complejidad computacional y no pueden ser ejecutadas en tiempo real empleando procesadores tradicionales. En cambio, como se demuestra en esta tesis, la paralelización de las distintas subtareas y su implementación sobre una GPU arrojan los resultados deseados de ejecución con tasas de refresco interactivas. Asimismo, se propone una técnica para la evaluación rápida de funciones de complejidad arbitraria especialmente indicada para su uso en una GPU. En primer lugar se estudia la aplicación de técnicas de síntesis de imágenes virtuales a partir de únicamente dos cámaras lejanas y no paralelas—en contraste con la configuración habitual en TV 3D de cámaras cercanas y paralelas—con información de color y profundidad. Empleando filtros de mediana modificados para la elaboración de un mapa de profundidad virtual y proyecciones inversas, se comprueba que estas técnicas son adecuadas para una libre elección del punto de vista. Además, se demuestra que la codificación de la información de profundidad con respecto a un sistema de referencia global es sumamente perjudicial y debería ser evitada. Por otro lado se propone un sistema de detección de objetos móviles basado en técnicas de estimación de densidad con funciones locales. Este tipo de técnicas es muy adecuada para el modelado de escenas complejas con fondos multimodales, pero ha recibido poco uso debido a su gran complejidad computacional. El sistema propuesto, implementado en tiempo real sobre una GPU, incluye propuestas para la estimación dinámica de los anchos de banda de las funciones locales, actualización selectiva del modelo de fondo, actualización de la posición de las muestras de referencia del modelo de primer plano empleando un filtro de partículas multirregión y selección automática de regiones de interés para reducir el coste computacional. Los resultados, evaluados sobre diversas bases de datos y comparados con otros algoritmos del estado del arte, demuestran la gran versatilidad y calidad de la propuesta. Finalmente se propone un método para la aproximación de funciones arbitrarias empleando funciones continuas lineales a tramos, especialmente indicada para su implementación en una GPU mediante el uso de las unidades de filtraje de texturas, normalmente no utilizadas para cómputo numérico. La propuesta incluye un riguroso análisis matemático del error cometido en la aproximación en función del número de muestras empleadas, así como un método para la obtención de una partición cuasióptima del dominio de la función para minimizar el error. ABSTRACT The evolution of smartphones, all equipped with digital cameras, is driving a growing demand for ever more complex applications that need to rely on real-time computer vision algorithms. However, video signals are only increasing in size, whereas the performance of single-core processors has somewhat stagnated in the past few years. Consequently, new computer vision algorithms will need to be parallel to run on multiple processors and be computationally scalable. One of the most promising classes of processors nowadays can be found in graphics processing units (GPU). These are devices offering a high parallelism degree, excellent numerical performance and increasing versatility, which makes them interesting to run scientific computations. In this thesis, we explore two computer vision applications with a high computational complexity that precludes them from running in real time on traditional uniprocessors. However, we show that by parallelizing subtasks and implementing them on a GPU, both applications attain their goals of running at interactive frame rates. In addition, we propose a technique for fast evaluation of arbitrarily complex functions, specially designed for GPU implementation. First, we explore the application of depth-image–based rendering techniques to the unusual configuration of two convergent, wide baseline cameras, in contrast to the usual configuration used in 3D TV, which are narrow baseline, parallel cameras. By using a backward mapping approach with a depth inpainting scheme based on median filters, we show that these techniques are adequate for free viewpoint video applications. In addition, we show that referring depth information to a global reference system is ill-advised and should be avoided. Then, we propose a background subtraction system based on kernel density estimation techniques. These techniques are very adequate for modelling complex scenes featuring multimodal backgrounds, but have not been so popular due to their huge computational and memory complexity. The proposed system, implemented in real time on a GPU, features novel proposals for dynamic kernel bandwidth estimation for the background model, selective update of the background model, update of the position of reference samples of the foreground model using a multi-region particle filter, and automatic selection of regions of interest to reduce computational cost. The results, evaluated on several databases and compared to other state-of-the-art algorithms, demonstrate the high quality and versatility of our proposal. Finally, we propose a general method for the approximation of arbitrarily complex functions using continuous piecewise linear functions, specially formulated for GPU implementation by leveraging their texture filtering units, normally unused for numerical computation. Our proposal features a rigorous mathematical analysis of the approximation error in function of the number of samples, as well as a method to obtain a suboptimal partition of the domain of the function to minimize approximation error.

Semantic Support for Computational Land-Use Modelling

Postprint

A formal framework for modelling complex network management systems

Society today is completely dependent on computer networks, the Internet and distributed systems, which place at our disposal the necessary services to perform our daily tasks. Subconsciously, we rely increasingly on network management systems. These systems allow us to, in general, maintain, manage, configure, scale, adapt, modify, edit, protect, and enhance the main distributed systems. Their role is secondary and is unknown and transparent to the users. They provide the necessary support to maintain the distributed systems whose services we use every day. If we do not consider network management systems during the development stage of distributed systems, then there could be serious consequences or even total failures in the development of the distributed system. It is necessary, therefore, to consider the management of the systems within the design of the distributed systems and to systematise their design to minimise the impact of network management in distributed systems projects. In this paper, we present a framework that allows the design of network management systems systematically. To accomplish this goal, formal modelling tools are used for modelling different views sequentially proposed of the same problem. These views cover all the aspects that are involved in the system; based on process definitions for identifying responsible and defining the involved agents to propose the deployment in a distributed architecture that is both feasible and appropriate.

Mathematical modelling of the lower urinary tract

The lower urinary tract is one of the most complex biological systems of the human body as it involved hydrodynamic properties of urine and muscle. Moreover, its complexity is increased to be managed by voluntary and involuntary neural systems. In this paper, a mathematical model of the lower urinary tract it is proposed as a preliminary study to better understand its functioning. Furthermore, another goal of that mathematical model proposal is to provide a basis for developing artificial control systems. Lower urinary tract is comprised of two interacting systems: the mechanical system and the neural regulator. The latter has the function of controlling the mechanical system to perform the voiding process. The results of the tests reproduce experimental data with high degree of accuracy. Also, these results indicate that simulations not only with healthy patients but also of patients with dysfunctions with neurological etiology present urodynamic curves very similar to those obtained in clinical studies.

Footwear bio-modelling: An industrial approach

There is a growing need within the footwear sector to customise the design of the last from which a specific footwear style is to be produced. This customisation is necessary for user comfort and health reasons, as the user needs to wear a suitable shoe. For this purpose, a relationship must be established between the user foot and the last with which the style will be made; up until now, no model has existed that integrates both elements. On the one hand, traditional customised footwear manufacturing techniques are based on purely artisanal procedures which make the process arduous and complex; on the other hand, geometric models proposed by different authors present the impossibility of implementing them in an industrial environment with limited resources for the acquisition of morphometric and structural data for the foot, apart from the fact that they do not prove to be sufficiently accurate given the non-similarity of the foot and last. In this paper, two interrelated geometric models are defined, the first, a bio-deformable foot model and the second, a deformable last model. The experiments completed show the goodness of the model, with it obtaining satisfactory results in terms of comfort, efficiency and precision, which make it viable for use in the sector.

Conceptual Modelling of Complex Network Management Systems

Society, as we know it today, is completely dependent on computer networks, Internet and distributed systems, which place at our disposal the necessary services to perform our daily tasks. Moreover, and unconsciously, all services and distributed systems require network management systems. These systems allow us to, in general, maintain, manage, configure, scale, adapt, modify, edit, protect or improve the main distributed systems. Their role is secondary and is unknown and transparent to the users. They provide the necessary support to maintain the distributed systems whose services we use every day. If we don’t consider network management systems during the development stage of main distributed systems, then there could be serious consequences or even total failures in the development of the distributed systems. It is necessary, therefore, to consider the management of the systems within the design of distributed systems and systematize their conception to minimize the impact of the management of networks within the project of distributed systems. In this paper, we present a formalization method of the conceptual modelling for design of a network management system through the use of formal modelling tools, thus allowing from the definition of processes to identify those responsible for these. Finally we will propose a use case to design a conceptual model intrusion detection system in network.

Modelling inpatient length of stay by a hierarchical mixture regression via the EM algorithm

The modelling of inpatient length of stay (LOS) has important implications in health care studies. Finite mixture distributions are usually used to model the heterogeneous LOS distribution, due to a certain proportion of patients sustaining-a longer stay. However, the morbidity data are collected from hospitals, observations clustered within the same hospital are often correlated. The generalized linear mixed model approach is adopted to accommodate the inherent correlation via unobservable random effects. An EM algorithm is developed to obtain residual maximum quasi-likelihood estimation. The proposed hierarchical mixture regression approach enables the identification and assessment of factors influencing the long-stay proportion and the LOS for the long-stay patient subgroup. A neonatal LOS data set is used for illustration, (C) 2003 Elsevier Science Ltd. All rights reserved.

SimuMap: a computational system for spatial modelling

The paper presents a computational system based upon formal principles to run spatial models for environmental processes. The simulator is named SimuMap because it is typically used to simulate spatial processes over a mapped representation of terrain. A model is formally represented in SimuMap as a set of coupled sub-models. The paper considers the situation where spatial processes operate at different time levels, but are still integrated. An example of such a situation commonly occurs in watershed hydrology where overland flow and stream channel flow have very different flow rates but are highly related as they are subject to the same terrain runoff processes. SimuMap is able to run a network of sub-models that express different time-space derivatives for water flow processes. Sub-models may be coded generically with a map algebra programming language that uses a surface data model. To address the problem of differing time levels in simulation, the paper: (i) reviews general approaches for numerical solvers, (ii) considers the constraints that need to be enforced to use more adaptive time steps in discrete time specified simulations, and (iii) scaling transfer rates in equations that use different time bases for time-space derivatives. A multistep scheme is proposed for SimuMap. This is presented along with a description of its visual programming interface, its modelling formalisms and future plans. (C) 2003 Elsevier Ltd. All rights reserved.

Applying ontologies to business and systems modelling techniques and perspectives: Lessons learned

For many years in the area of business systems analysis and design, practitioners and researchers alike have been searching for some comprehensive basis on which to evaluate, compare, and engineer techniques that are promoted for use in the modelling of systems' requirements. To date, while many frameworks, factors, and facets have been forthcoming, none appear to be based on a sound theory. In light of this dilemma, over the last 10 years, attention has been devoted by researchers to the use of ontology to provide some theoretical basis for the advancement of the business systems modelling discipline. This paper outlines how we have used a particular ontology for this purpose over the last five years. In particular we have learned that the understandability and the applicability of the selected ontology must be clear for IS professionals, the results of any ontological evaluation must be tempered by economic efficiency considerations of the stakeholders involved, and ontologies may have to be focused for the business purpose and type of user involved in the modelling situation.

Numerical simulation of stochastic ordinary differential equations in biomathematical modelling

In this work we discuss the effects of white and coloured noise perturbations on the parameters of a mathematical model of bacteriophage infection introduced by Beretta and Kuang in [Math. Biosc. 149 (1998) 57]. We numerically simulate the strong solutions of the resulting systems of stochastic ordinary differential equations (SDEs), with respect to the global error, by means of numerical methods of both Euler-Taylor expansion and stochastic Runge-Kutta type. (C) 2003 IMACS. Published by Elsevier B.V. All rights reserved.

Process systems modelling and applications in granulation: A review

Granulation is one of the fundamental operations in particulate processing and has a very ancient history and widespread use. Much fundamental particle science has occurred in the last two decades to help understand the underlying phenomena. Yet, until recently the development of granulation systems was mostly based on popular practice. The use of process systems approaches to the integrated understanding of these operations is providing improved insight into the complex nature of the processes. Improved mathematical representations, new solution techniques and the application of the models to industrial processes are yielding better designs, improved optimisation and tighter control of these systems. The parallel development of advanced instrumentation and the use of inferential approaches provide real-time access to system parameters necessary for improvements in operation. The use of advanced models to help develop real-time plant diagnostic systems provides further evidence of the utility of process system approaches to granulation processes. This paper highlights some of those aspects of granulation. (c) 2005 Elsevier Ltd. All rights reserved.

Population-based continuous optimization, probabilistic modelling and mean shift

Evolutionary algorithms perform optimization using a population of sample solution points. An interesting development has been to view population-based optimization as the process of evolving an explicit, probabilistic model of the search space. This paper investigates a formal basis for continuous, population-based optimization in terms of a stochastic gradient descent on the Kullback-Leibler divergence between the model probability density and the objective function, represented as an unknown density of assumed form. This leads to an update rule that is related and compared with previous theoretical work, a continuous version of the population-based incremental learning algorithm, and the generalized mean shift clustering framework. Experimental results are presented that demonstrate the dynamics of the new algorithm on a set of simple test problems.

Modelling sediment delivery ratio over the Murray Darling Basin

This paper presents a scientific and technical description of the modelling framework and the main results of modelling the long-term average sediment delivery at hillslope to medium-scale catchments over the entire Murray Darling Basin (MDB). A theoretical development that relates long-term averaged sediment delivery to the statistics of rainfall and catchment parameters is presented. The derived flood frequency approach was adapted to investigate the problem of regionalization of the sediment delivery ratio (SDR) across the Basin. SDR, a measure of catchment response to the upland erosion rate, was modeled by two lumped linear stores arranged in series: hillslope transport to the nearest streams and flow routing in the channel network. The theory shows that the ratio of catchment sediment residence time (SRT) to average effective rainfall duration is the most important control in the sediment delivery processes. In this study, catchment SRTs were estimated using travel time for overland flow multiplied by an enlargement factor which is a function of particle size. Rainfall intensity and effective duration statistics were regionalized by using long-term measurements from 195 pluviograph sites within and around the Basin. Finally, the model was implemented across the MDB by using spatially distributed soil, vegetation, topographical and land use properties under Geographic Information System (GIs) environment. The results predict strong variations in SDR from close to 0 in floodplains to 70% in the eastern uplands of the Basin. (c) 2005 Elsevier Ltd. All rights reserved.

Modelling trickle irrigation: Comparison of analytical and numerical models for estimation of wetting front position with time

Irrigation practices that are profligate in their use of water have come under closer scrutiny by water managers and the public. Trickle irrigation has the propensity to increase water use efficiency but only if the system is designed to meet the soil and plant conditions. Recently we have provided a software tool, WetUp (http://www.clw.csiro.au/products/wetup/), to calculate the wetting patterns from trickle irrigation emitters. WetUp uses an analytical solution to calculate the wetted perimeter for both buried and surface emitters. This analytical solution has a number of assumptions, two of which are that the wetting front is defined by water content at which the hydraulic conductivity (K) is I mm day(-1) and that the flow occurs from a point source. Here we compare the wetting patterns calculated with a 2-dimensional numerical model, HYDRUS2D, for solving the water flow into typical soils with the analytical solution. The results show that the wetting patterns are similar, except when the soil properties result in the assumption of a point source no longer being a good description of the flow regime. Difficulties were also experienced with getting stable solutions with HYDRUS2D for soils with low hydraulic conductivities. (c) 2005 Elsevier Ltd. All rights reserved.

Data flow and validation in workflow modelling

A complete workflow specification requires careful integration of many different process characteristics. Decisions must be made as to the definitions of individual activities, their scope, the order of execution that maintains the overall business process logic, the rules governing the discipline of work list scheduling to performers, identification of time constraints and more. The goal of this paper is to address an important issue in workflows modelling and specification, which is data flow, its modelling, specification and validation. Researchers have neglected this dimension of process analysis for some time, mainly focussing on structural considerations with limited verification checks. In this paper, we identify and justify the importance of data modelling in overall workflows specification and verification. We illustrate and define several potential data flow problems that, if not detected prior to workflow deployment may prevent the process from correct execution, execute process on inconsistent data or even lead to process suspension. A discussion on essential requirements of the workflow data model in order to support data validation is also given..