This thesis presents a real-time multi-face tracking system, which is able to track multiple faces for live videos, broadcast, real-time conference recording, etc. The real-time output is one of the most significant advantages. Our proposed tracking system is comprised of three parts: face detection, feature extraction and tracking. We deploy a three-layer Convolutional Neural Network (CNN) to detect a face, a one-layer CNN to extract the features of a detected face and a shallow network for face tracking based on the extracted feature maps of the face. The performance of our multi-face tracking system enables the tracker to run in real-time without any on-line training. This algorithm does not need to change any parameters according to different input video conditions, and the runtime cost will not be affected significantly by an the increase in the number of faces being tracked. In addition, our proposed tracker can overcome most of the generally difficult tracking conditions which include video containing a camera cut, face occlusion, false positive face detection, false negative face detection, e.g. due to faces at the image boundary or faces shown in profile. We use two commonly used metrics to evaluate the performance of our multi-face tracking system demonstrating that our system achieves accurate results. Our multi-face tracker achieves an average runtime cost around 0.035s with GPU acceleration and this runtime cost is close to stable even if the number of tracked faces increases. All the evaluation results and comparisons are tested with four commonly used video data sets.

L’informatique multi-nuages s’est imposée comme un paradigme de choix pour créer des applications distribuées à large échelle s’exécutant à des emplacements géographiques répartis. L’informatique multi-nuages consiste en l’utilisation de multiples environnements de nuages indépendants qui ne nécessitent pas d’accord a priori entre les fournisseurs de nuage ou un tiers. Toutefois, ces applications conçues pour un environnement multi-nuages doivent faire face à de véritables défis en terme d’architecture, de modèle et de technologies. L’utilisation de l’informatique multi-nuages se heurte à l’hétérogénéité et à la complexité des offres de nuage. Ainsi, l’informatique multi-nuages doit faire face aux défis de la portabilité, de l’approvisionnement, de l’élasticité et de la haute disponibilité que nous identifions dans cette thèse.Dans ce travail de thèse, nous proposons un modèle d’applications nommé soCloud qui adresse ces quatre défis. C’est un modèle basé sur le standard SCA du consortium OASIS pour concevoir de manière simple et cohérente des applications distribuées à large échelle pour un environnement multi-nuages. Un nouveau langage dédié d’élasticité a été proposé pour exprimer efficacement l’élasticité d’applications multi-nuages par l’abstraction. Nous proposons aussi une plateforme multi-nuages soCloud conçue pour déployer, exécuter et gérer des applications réparties à travers plusieurs nuages. Le modèle d’applications soCloud a été utilisé pour la mise en œuvre de trois applications distribuées déployées dans un environnement multi-nuages. Quant à la plateforme soCloud, elle a été implantée, déployée et expérimentée sur dix nuages. / Multi-cloud computing has established itself as a paradigm of choice for creating very large scale world wide distributed applications. Multi-cloud computing is the usage of multiple, independent cloud environments, which assumed no priori agreement between cloud providers or third party. However, these applications, designed for multi-cloud envi-ronments, have to face real challenges in term of design, architecture, and technology. The possibility of using multi-cloud faces the heterogeneity and complexity of cloud solutions.Thus, multi-cloud computing has to face several challenges such as portability, provisioning, elasticity, and high availability we have identified in this thesis. In this thesis, we propose soCloud both amodel and a platformthat tackle these four chal-lenges. This model is based on the OASIS Service Component Architecture (SCA) standard to design distributed large scale applications for multi-cloud environments. Anew language is proposed to effectively express the elasticity of multi-cloud applications through abstrac-tion. The multi-cloud platform is designed to deploy and manage distributed applications across multi-clouds. The soCloud model is illustrated on three distributed applications deployed in multi-cloud environments. The soCloud platform has been implemented, deployed and experi-mented on top of ten existing cloud providers.These experiments are used to validate the novelty of the contributed solutions. With our contributions, we aim to provide a simple and effective way to design, deploy, run, and manage distributed applications for a multi-cloud environment by proposing a model and platform.

Le développement de nouveaux matériaux nécessite la détermination de leurs propriétés mécaniques aux échelles macro, micro et nanométrique. Parmi les méthodes expérimentales les plus courantes, l’essai d’indentation instrumentée permettant de déterminer le module d’élasticité et la dureté du matériau à toutes ces échelles de mesure est sans doute le plus utilisé. Cependant, les valeurs obtenues en micro et en nanoindentation sont souvent en apparente contradiction. Cela a conduit de nombreux auteurs à s’interroger sur la validité des mesures expérimentales, sur l’identification des processus physiques mis en jeu lors de l’enfoncement de l’indenteur ou encore sur la pertinence des propriétés mesurées et des échelles de mesure. Sans prétendre étudier tous les aspects de manière exhaustive ce problème complexe, nous proposons cependant quelques pistes de réflexions en particulier sur le passage nano/micro en dureté. Sur la base de la théorie du gradient de plasticité développé par Nix et Gao, nous définissons un facteur d’échelle qui relie certaines données expérimentales à des propriétés fondamentales du matériau. Nous montrons également que ce facteur d’échelle, capable de caractériser la résistance à la déformation plastique, peut donner des informations complémentaires pour étudier le comportement des matériau soumis à différentes conditions d’indentation. Dans une deuxième partie, nous étudions le fluage par indentation comme une alternative aux essais classiques de fluage qui durent généralement plusieurs mois. Même si le résultat ne répond pas complètement aux objectifs fixés, nous montrons qu’un modèle rhéologique adapté permet de retrouver le module d’élasticité et de calculer un coefficient de viscosité à température ambiante du matériau. / Development of new materials often requires the determination of their mechanical properties at macro, micro and nanometric scales. Among the usual experimental methods, the depth-sensing indentation test allowing determining the elastic modulus and the hardness of the material over all the scales of measurement is probably the most employed. However, the values obtained in nano and microindentation ranges are in apparent contradiction. As a consequence, numerous authors have some questioning on the validity of the experimental data, on the identification of the physical phenomenon taking place during the indenter penetration, on the pertinence of the mechanical properties and on the scales of measurement. Without pretending to study exhaustively all the aspects of this complex problem, we can propose some paths for further research in particular on the comparison between nano and microhardness. Starting from the strain gradient plasticity theory developed by Nix and Gao, we defined a hardness length-scale factor which allows connecting some experimental data to fundamental properties of the material. We also show that this factor, able to represent the plastic deformation resistance, can give additional information to study the mechanical behavior of materials submitted to various indentation conditions. In a second part, we study creep by indentation as an alternative method to the usual creep test which generally requires some months of testing. Even if the result does not agree well with this objective, we show that a rheological model can be helpful to determine the elastic modulus and to calculate the coefficient of viscosity at room temperature of the material.

Ce travail, réalisé dans le cadre d'un contrat CIFRE entre le groupe Volvo, le LAGIS et le LM2S, traite des problèmes de planification des tâches de maintenance sur des véhicules industriels. Le but est de proposer un service de planification de maintenance adaptée au véhicule spécifique avec un coût de maintenance optimisé.Pour un système multi-composant, deux ensembles de composants peuvent être définis suivant le type d'information disponible concernant leurs états: âge ou niveau de dégradation. Sur la base de ces deux ensembles, des dates optimales de maintenance pour chaque composant peuvent être obtenues en minimisant une fonction de coût qui tient compte des coûts préventif, correctif et d'inspection éventuelle et qui dépend de la probabilité de panne. Dans le cas des composants à maintenance basée sur l'âge, la fonction de probabilité de panne est fixée a priori ; dans le cas des composants à détérioration elle est recalculée à chaque inspection.Les opérations de maintenance sont ensuite regroupées sur un horizon roulant. Certaines opérations doivent être déplacées par rapport à leur date optimale ce qui entraîne l'ajout d'un coût de pénalité. Les regroupements sont effectués en tenant compte des coûts de logistique des arrêts de maintenance, des coûts de pénalité de chaque opération et d'un intervalle de temps minimal entre deux groupes de maintenances qui correspond au temps de préparation de maintenance. La planification est obtenue en maximisant la différence entre la somme des coûts des maintenances individuelles et le coût global en tenant compte des groupes de tâches de maintenance. Le planning de maintenance est remis à jour à la prochaine inspection. / This thesis deals with the maintenance operation scheduling issues of commercial vehicles. It is a CIFRE collaboration between the Volvo group, LAGIS and LM2S laboratories. The aim is to suggest a maintenance scheduling service adapted to the vehicle with an optimized maintenance cost.The vehicle is considered as a multi-component system. For each component, the optimal maintenance dates are computed using a maintenance cost function. This cost function depends on the preventive cost, the corrective cost and possibly the inspection cost, and also on the component failure probability function. In our system, two types of components are defined according to the available component state information : age or degradation level. The failure probability function is fixed a priori for the first component class while it is updated at each inspection for deteriorating and monitored components.The maintenance operations are grouped on a rolling horizon. Some operations must be shifted from their optimal maintenance dates, which implies a penalty cost. The maintenance scheduling depends on the set-up cost, the penalty cost of each operation, and a minimal time interval between two maintenance groups which represents the preparation time. The set-up costs depend on the number of maintenance stops. The saving cost is computed from the difference between the maintenance cost with individual operations and the maintenance cost with the defined operation groups. The developed algorithm provides the optimal maintenance planning according to the available degradation data and the operational constraints. Maintenance scheduling is updated at the next inspection date.

La voiture est le mode de transport le plus utilisé en Europe et en Amérique du Nord. Aujourd’hui, il est de plus en plus sécurisé notamment grâce à des systèmes d’assistance à la conduite. Toutefois, il représente encore la plus grande part d’accidents sur la route en France dont 90 % seraient de cause humaine. A partir de 2020, de nouveaux types de véhicules vont apparaître sur la route : ce sont des véhicules dont la décision ne dépendra plus uniquement du conducteur humain, appelés couramment véhicules autonomes.Pour concevoir de tels systèmes, trois grandes exigences doivent être respectées à la fois : la sécurité (respect du code de la route), l’efficacité (aller le plus vite possible) et le confort (ne pas se sentir en danger dans le véhicule). En plus d’imaginer un véhicule où seuls les capteurs embarqués fournissent les informations nécessaires pour conduire, on peut ajouter la possibilité de communiquer avec d’autres véhicules ou l’infrastructure de la route. Ce dernier point doit prendre en compte une quatrième exigence, la courtoisie (ne pas avoir un impact négatif sur les véhicules environnants). Il peut permettre également de gérer une situation proche où les véhicules conduits par des êtres humains et les véhicules autonomes vont se trouver dans un même environnement.Pour étudier ces comportements, la simulation informatique peut être une bonne solution pour mettre en place un éventail de scénarios possibles dans différents environnements. Toutefois, cela impose un niveau d’abstraction pouvant affecter le niveau de réalisme du modèle.Dans ce manuscrit, nous avons défini des méthodes et outils afin de définir une méthodologie permettant de concevoir des Systèmes de Transport Intelligents Coopératifs. A partir d'un modèle de véhicule, nous allons le simuler, puis le vérifier formellement. A partir du modèle obtenu, on assure une compatibilité avec un simulateur plus réaliste. De plus, à partir d'un outil de simulation, nous avons mis en place un protocole de coopération permettant aux véhicules de s’adapter plus facilement à des environnements routiers actuels. / The car is the most used mode of transport in Europe and North America. Today, it is increasingly secure thanks to driver assistance systems. However, it is still the largest share of road accidents in France, 90 % of which are caused by humans. From 2020, new types of vehicles will appear on the road: they are vehicles whose decision will no longer depend only on the human driver, commonly called autonomous vehicles.To design such systems, three main requirements must be respected at the same time: the safety (respect of traffic laws), the efficiency (go as fast as possible) and the comfort (not to feel in danger in the vehicle). In addition to imagine a vehicle where only on-board sensors provide the necessary information to drive, one can add the ability to communicate with other vehicles or the road infrastructure. This last point have to take into account a fourth requirement, courtesy (does not have a negative impact on surrounding vehicles). It can also help to manage a similar situation where vehicles driven by human beings and autonomous vehicles will be in the same environment.To study these behaviors, computer simulation can be a good solution to set up a range of possible scenarios in different environments. However, this imposes a level of abstraction that can affect the level of realism of the model.In this PhD thesis, we defined methods and tools to define a methodology for designing Cooperative Intelligent Transport Systems. From a vehicle model, we have shown that we can simulate it and then formally check it. From the model obtained, compatibility with a more realistic simulator is ensured. In addition, based on a simulation tool, we have implemented a cooperation protocol that allows vehicles to adapt more easily to current road environments.

This thesis describes a system to implement a distributed multi-user process interface to allow the PDP-11/23 computer in the Electrical Engineering department at UCT to be used for process control. The use of this system is to be shared between postgraduate students for research and undergraduates for doing real-time control projects. The interface may be used concurrently by several users, and access is controlled in such a way as to prevent users' programs from interfering with one another. The process interface hardware used was a GEC Micro-Media system, which is a stand-alone process interface system communicating with a host (the PDP-11/23) via a serial line. Hardware to drive a 600-metre serial link at 9600 baud between the PDP-11/23 and the Media interface was designed and built. The software system on the host, written in RTL/2, holds-all data from the interface in a resident common data-base and continually updates it. Access to the interface by applications programs is done indirectly by reading and writing to the database, for which purpose a library of user interface routines is provided. To allow future expansion and modification of the Media interface, software (also written in RTL/2) for an LSI-11 minicomputer interfaced to the Media bus was developed which emulates the operation of the GEC proprietary Micro-Media software. A program to download this software into the LSI-11 was written. A suite of diagnostic programs enables testing of the system hardware and software at various levels. To ease testing, teaching, and applications programming, a general-purpose simulation package for the simulation of analogue systems was developed, as well as graphics routines for use with a Tektronix 4010 plotting terminal. A. real-time computing project for a class of undergraduates was run in 1983. This project made extensive use of the system and demonstrated its viability.

This dissertation concerns with the formulation of an improved multi-target multi-Bernoulli (MeMBer) filter and the use of the joint multi-target (JoM) estimator in an effective and efficient manner for a specific implementation of MeMBer filters. After reviewing random finite set (RFS) formalism for multi-target tracking problems and the related Bayes estimators the major contributions of this dissertation are explained in detail. The second chapter of this dissertation is dedicated to the analysis of the relationship between the multi-Bernoulli RFS distribution and the MeMBer corrector. This analysis leads to the formulation of an unbiased MeMBer filter without making any limiting assumption. Hence, as opposed to the popular cardinality balanced multi-target multi-Bernoulli (CBMeMBer) filter, the proposed MeMBer filter can be employed under the cases when sensor detection probability is moderate to low. Furthermore, a statistical refinement process is introduced to improve the stability of the estimated cardinality of targets obtained from the proposed MeMBer filter. The results from simulations demonstrate the effectiveness of the improved MeMBer filter. In Chapters III and IV, the Bayesian optimal estimators proposed for the RFS based multi-target tracking filters are examined in detail. First, an optimal solution to the unknown constant in the definition of the JoM estimator is determined by solving a multi-objective optimization problem. Thus, the JoM estimator can be implemented for tracking of a Bernoulli target using the optimal joint target detection and tracking (JoTT) filter. The results from simulations confirm assertions about its performance obtained by theoretical analysis in the literature. Finally, in the third chapter of this dissertation, the proposed JoM estimator is reformulated for RFS multi-Bernoulli distributions. Hence, an effective and efficient implementation of the JoM estimator is proposed for the Gaussian mixture implementations of the MeMBer filters. Simulation results demonstrate the robustness of the proposed JoM estimator under low-observable conditions. / Thesis / Doctor of Philosophy (PhD)

The thought of building on this land is frightening to me. Construction is necessarily accompanied by a relative level of destruction, and I always have to question whether the positive outweighs the negative. It is rare that the act of building is approached with the level of gravity it deserves, considering the significance of what is taking place. It is the replacement and/or displacement of "what is" with what we think "should be". And what "should be"? Luigi Snozzi tells us to "Destroy with sensibility." So the question becomes, what is sensible, and conversely (and maybe more importantly), what is insensible. Architecture has the role of establishing the relationship that will exist between inhabitants and inhabited before the former even arrive. In a building that is designed with careful consideration of its surroundings, the inhabitants are themselves likely to extend a more careful consideration toward their environment. It is a fortunate relationship for both, as the inhabited is treated to at least a greater level of regard if not reverence, while the inhabitants benefit from the richness of experience that comes with a positive connection to place. In buildings where this role is neglected, the unfortunate relationship established by default is one of dominion with disregard. / Master of Architecture

Multi-dimensional functions are characterized by a large number of parameters on which the functional value depends. They are commonly used in engineering problems such as image analysis and system modeling. Multi-dimensional function spaces are very difficult to understand due to their multi-dimensional nature and the presence of a large number of data points in the functional space. A point called the focal point is selected by the user in the vast multi-dimensional parameter space. Rays called "focal rays" emanate from the focal point in all directions to the boundaries of the functional space. The focal rays contain functional data points. The focal point is mapped onto the center of the visualization with the focal rays arranged radially around it. The degree of detail decreases as we move away from the focal point toward the edges of the visualization in accordance with the focus+context technique. The functional values are mapped onto a color scheme with shades of green representing positive function values, and shades of red representing negative function values. Interactive features such as the ability to change the focal point, highlighting of functional values aid the user in exploring and analyzing the functional space. The algorithm, practical applications of the visualization approach and results of formative evaluation studies are also elaborated in this thesis. The contributions of this thesis are fourfold, namely, providing an overview of the functional space, equal treatment of all dimensions, improved scalability and a smooth blending of details with the overview. / Master of Science

We present an algorithm to explore an orthogonal polygon using a team of p robots. This algorithm combines ideas from information-theoretic exploration algorithms and computational geometry based exploration algorithms. The algorithm is based on a single-robot polygon exploration algorithm and a tree exploration algorithm. We show that the exploration time of our algorithm is competitive (as a function of p) with respect to the offline optimal exploration algorithm. We discuss how this strategy can be adapted to real-world settings to deal with noisy sensors. In addition to theoretical analysis, we investigate the performance of our algorithm through simulations for multiple robots and experiments with a single robot. / Master of Science / In applications such as disaster recovery, the layout of the environment is generally unknown. Hence, there is a need to explore the environment in order to effectively perform search and rescue. Exploration of unknown environments using a single robot is a well studied problem. We present an algorithm to perform the task with a team of p robots for the specific case of orthogonal polygons, i.e. polygonal environments where each side is aligned with either the X or the Y axis. The algorithm is based on a single-robot polygon exploration algorithm and a tree exploration algorithm. We show that the exploration time of our algorithm is competitive (as a function of p) with respect to the optimal offline algorithm. We then optimize the information gain of the path followed by the robots by allowing local detours in order to decrease the entropy in the map.