Thesis (Ph.D.)--University of Nebraska-Lincoln, 2008. / Title from title screen (site viewed Nov. 25, 2008). PDF text: xiii, 173 p. : ill. ; 2 Mb. UMI publication number: AAT 3315261. Includes bibliographical references. Also available in microfilm and microfiche formats.

Context-aware computing enables mobile applications to discover and benefit from valuable context information, such as user location, time of day and current activity. However, determining the users’ context throughout their daily activities is one of the main challenges of context-aware computing. With the increasing number of built-in mobile sensors and other input sources, existing context models do not effectively handle context information related to personal user context. The objective of this research was to develop an improved context-aware model to support the context awareness needs of mobile applications. An existing context-aware model was selected as the most complete model to use as a basis for the proposed model to support context awareness in mobile applications. The existing context-aware model was modified to address the shortcomings of existing models in dealing with context information related to personal user context. The proposed model supports four different context dimensions, namely Physical, User Activity, Health and User Preferences. A prototype, called CoPro was developed, based on the proposed model, to demonstrate the effectiveness of the model. Several experiments were designed and conducted to determine if CoPro was effective, reliable and capable. CoPro was considered effective as it produced low-level context as well as inferred context. The reliability of the model was confirmed by evaluating CoPro using Quality of Context (QoC) metrics such as Accuracy, Freshness, Certainty and Completeness. CoPro was also found to be capable of dealing with the limitations of the mobile computing platform such as limited processing power. The research determined that the proposed context-aware model can be used to successfully support context awareness in mobile applications. Design recommendations were proposed and future work will involve converting the CoPro prototype into middleware in the form of an API to provide easier access to context awareness support in mobile applications.

Online information access on mobile devices is increasing as a result of the growth in the use of Internet-enabled handheld (or pocket-size) devices. The combined influence of recent enabling technologies such as Web 2.0, mobile app stores and improved wireless networks have driven the increase in online applications that allow users to access various types of information on mobile devices regardless of time and location. Examples of such applications (usually shortened to app) include: social media, such as FacebookTM App and TwitterTM App, banking applications such as (Standard Bank South Africa)TM Mobile Banking App and First National Bank (FNB) BankingTM App, and news application such as news 24TM App and BBCTM News App. Online businesses involved in buying, selling and business transaction processing activities via the Internet have exploited the opportunity to extend electronic commerce (e-commerce) initiatives into mobile commerce (m-commerce). Online businesses that interact with end user customers implement business to consumer (B2C) m-commerce applications that enable customers to access and browse product catalogue information on mobile devices, anytime, anywhere. Customers accessing electronic product catalogue information on a mobile device face a number of challenges such as a long list of products presented on a small screen and a longer information download time. These challenges mainly originate from the limiting and dynamic nature of the mobile apps operating environment, for example, dynamic location, bandwidth fluctuations and, diverse and limited device features, collectively referred to as context. The goal of this research was to design and implement a context-aware model that can be incorporated into an m-commerce application in order to improve the presentation of product catalogue information on m-commerce storefronts. The motivation for selecting product catalogue is prompted by literature which indicates that improved presentation of information in m-commerce (and e-commerce) applications has a positive impact on usability of the websites. Usable m-commerce (and e-commerce) websites improve efficiency in consumer behaviour that impacts sales, profits and business growth. The context-aware model aimed at collecting context information within the user environment and utilising it to determine optimal retrieval and presentation of product catalogue in m-commerce. An integrated logical context sensor and Mathematical algorithms were implemented in the context-aware model. The integrated logical context sensor was responsible for the collection of different types of predetermined context information such as device specification or capabilities, connection bandwidth, location and time of the day as well as the user profile. The algorithms transformed the collected context information into usable formats and enabled optimal retrieval and presentation of product catalogue data on a specific mobile device. Open-source implementation tools were utilised to implement components of the model including: HTML5, PhP, JavaScript and MySQL database. The context-aware model was incorporated into an existing m-commerce application. Two user evaluation studies were conducted during the course of the research. The first evaluation was to evaluate the accuracy of information collected by the context sensor component of the model. This survey was conducted with a sample of 30 users from different countries across the world. In-between the context sensor and main evaluation surveys, a pilot study was conducted with a sample of 19 users with great experience in mobile application development and use from SAP Next Business and Technology, Africa. Finally an overall user evaluation study was conducted with a sample of 30 users from a remote area called Kgautswane in Limpopo Province, South Africa. The results obtained indicate that the context-aware model was able to determine accurate context information in real-time and effectively determine how much product information should be retrieved and how the information should be presented on a mobile device interface. Two main contributions emerged from the research, first the research contributed to the field of mobile Human Computer Interaction. During the research, techniques of evaluating and improving usability of mobile applications were demonstrated. Secondly, the research made a significant contribution to the upcoming field of context-aware computing. The research brought clarity with regard to context-aware computing which is lacking in existing, current research despite the field’s proven impact of improving usability of applications. Researchers can utilise contributions made in this research to develop further techniques and usable context-aware solutions.

Dense stereo matching is one of the most extensively investigated topics in computer vision, since it plays an important role in many applications such as 3D scene reconstruction. In this thesis, a novel dense stereo matching method is proposed based on edge-aware truncated minimum spanning tree (T-MST). Instead of employing non-local cost aggregation on traditional MST which is only generated from color differences of neighbouring pixels, a new tree structure, "Edge-Aware T-MST", is proposed to aggregate the cost according to the image texture. Specifically, cost aggregations are strongly enforced in large planar textureless regions due to the truncated edge weights. Meanwhile, the "edge fatten" effect is suppressed by employing a novel hybrid edge-prior which combines edge-prior and superpixel-prior to locate the potential disparity edges. Then a widely used Winner-Takes-All (WTA) strategy is performed to establish initial disparity map. An adaptive non-local refinement is also performed based on the stability of initial disparity estimation. Given the stereo images from Middlebury benchmark, we estimate the disparity maps by using our proposed method and other five state-of-the-art tree-based non-local matching methods. The experimental results show that the proposed method successfully produced reliable disparity values within large planar textureless regions and around object disparity boundaries. Performance comparisons demonstrate that our proposed non-local stereo matching method based on edge-aware T-MST outperforms current non-local tree-based state-of-the-art stereo matching methods in most cases, especially in large textureless planar regions and around disparity bounaries.

Context information available in wearable devices is believed to be useful in many ways. It allows for hiding much of the complexity from the user, thus enabling simpler user interfaces and less user interaction when carrying out tasks on behalf of a user, as well as enabling network operators to provide a better interface to thirdparty service providers who will provide and deliver wireless services. Using the available context information from the wearable device, optimization of service delivery in wireless networks, such as setting up optimal delivery paths between two wearable devices, may be possible without using a third party to do negotiations. In order to fully enable context-awareness, a clear model for how to sense, manage, derive, store, and exchange context information must be defined. This will then provide the platform needed to enable development of context-aware applications that can exploit the possibilities of context-aware computing. The model must take into consideration parameters such as memory usage and power and bandwidth consumption, in order to be efficient on all types of platforms and in all types of networks. It must also be modular enough to survive replacing and upgrading of internal parts. Today little research is available about sensing context information, sensor management, APIs towards other applications, and how and how often to present context information to applications. Since context aware computing relies heavily on the ability to obtain and represent context information, sensing strategies greatly affect efficiency and performance. It is therefore of great interest to develop and evaluate models for carrying out these tasks in order to exploit the results of context awareness research. This thesis will identify and design several components of such a model, as well as test and evaluate the design, in order to be able to make conclusions to whether is lives up to the expectations stated. In order to make the proper design decisions, a full understanding of the context-awareness research area and the goals and purposes of context-aware computing are required. To understand the entire picture is crucial to find a suitable solution. Therefore, determining an efficient sensor input and management strategy, along with a powerful and flexible API for applications, which are the goals of this thesis, fully qualifies as a significant master thesis assignment. / Information om bärbara enheters omgivning som kan göras tillgänglig i enheten, tros kunna vara användbart på många sätt. Det kan möjliggöra gömmande av komplexitet från användaren, vilket ger enklare användargränssnitt och mindre användarinteraktivitet, när utförandet av uppdrag från användaren sker, eller underlätta för en nätverksoperatör som tillhandahåller ett bättre gränssnitt gentemot en tredje part, som tillhandahåller och levererar trådlösa tjänster. Genom att utnyttja tillgänglig information om omgivningen från en bärbar enhet, kan man optimera leverans av tjänster i trådlösa nätverk, så som att hitta en optimal kommunikationsväg mellan två bärbara enheter, utan att använda sig av förhandlingar med en tredje part. För att till fullo möjliggöra ett sådant omgivningsmedvetande, krävs en tydlig modell för att uppfatta, förfina, lagra och utbyta det data som beskriver omgivningen. Denna modell kan då utgöra en plattform som möjliggör utveckling av omgivningsmedvetande applikationer, som kan utnyttja och reagera på de data som beskriver omgivningen. Modellen måste ta hänsyn till parametrar så som minneskonsumtion och batteri- och bandbreddsförbrukning, för att vara effektiv på alla typer av plattformar och i alla typer av nätverk. Den måste också bestå av tillräckligt väl separerade moduler för att klara av byten och uppgraderingar av dess beståndsdelar. Idag finns endast lite tillgänglig forskning om insamlandet av omgivningsdata, hanteringen av sensorer, gränssnitt gentemot mot applikationer och hur och hur ofta omgivningsdata skall presenteras för applikationer. Eftersom omgivningsmedvetenhet beror av möjligheten att införskaffa och representera omgivningsdata, påverkar strategier för att uppfatta omgivningen både effektivitet och prestanda. Det finns därför ett stort intresse i att utveckla och utvärdera modeller för utförandet av dessa uppdrag och för att utforska forskningsresultat om omgivningsmedvetande. Denna rapport identifierar och konstruerar flera komponenter till en sådan modell, samt testar och utvärderar denna för att kunna dra slutsatser om huruvida den lever upp till de förväntningar som finns. För att kunna göra en fullgod konstruktion, krävs en ingående förståelse i forskningsområdet omgivningsmedvetande och syften och mål med densamma. Att förstå den övergripande bilden är nyckeln till en passande lösning. Konstruktion av effektiva strategier för att uppfatta omgivningen, tillsammans med ett kraftfullt och flexibelt API gentemot applikationer, vilket är målen med denna rapport, kvalificerar sig därför som ett examensarbete.

Stress is now recognized as one of the major causes of physical and psychological illness. It is known as a reaction to surrounding environmental threats and the best way to manage it is to understand its triggers. Although people continuously react to their surrounding environments, they sometimes are not aware that certain elements in their environment are considered to be stressors. Based on this fact, researchers have recently proposed context-aware stress management systems. Most of the proposed systems use context data to provide real time stress monitoring and visualization, along with intervention techniques. However, these interventions are limited to the second and tertiary stages and very little attention has been given to the primary stage. In this thesis, we introduce a system called CASP. The system’s objective is to provide stress status predictions based on a user’s current contextual data. Therefore, a detection method is developed using heart rate variability (HRV) as a stress indicator to deliver personalized context-aware stress reports. Based on the predicted status, the system provides users with stress interventions at an early stage in order to help avoid and/or eliminate the occurrence of stress. Our evaluation results show that the CASP system is able to predict the stress status of a user with an averaged accuracy of 78.23% through our limited activity, when compare to a stress status measured using physiological signals. Moreover, it provides prediction models that adapt to the changing nature of both the user’s stress status and the surrounding environment.

Que le phénomène découle d’une prise de conscience des conséquences sur l’environnement, d’une opportunité économique ou d’une question de réputation et de commerce, la réduction des émissions de gaz à effets de serre est récemment devenue un objectif de premier plan. Les individus, les entreprises et les gouvernements effectuent un effort important pour réduire la dépense énergétique de multiples secteurs d’activité. Parallèlement, les technologies de l’information et de la communication sont de plus en plus présentes dans la plupart des activités humaines et l’on a estimé que 2% des émissions de gaz à effets de serre pouvaient leur être attribuées, cette proportion atteignant 10 % dans les pays fortement industrialisés [1, 2]. Si ces chiffres paraissent raisonnables aujourd’hui, ils sont certainement appelés à croître à l’avenir. À l’heure du cloud computing, les infrastructures de calcul et de communication demandent de plus en plus de performance et de disponibilité et imposent l’utilisation de matériels puissants et engendrant une consommation d’énergie importante du fait de leur fonctionnement direct, mais aussi à cause du refroidissement qu’ils nécessitent. En outre, les contraintes de disponibilité imposent une conception d’architectures redondantes et dimensionnées sur une charge crête. Les infrastructures sont donc souvent sous-utilisées et adapter leur niveau de performance à la charge effectivement constatée constitue une piste d’optimisation prometteuse à divers niveaux. Si l’on adopte un strict point de vue environnemental, l’objectif du Green Networking consiste à réduire le volume d’émissions de gaz à effets de serre dues au processus de communication. L’utilisation de sources d’énergie renouvelables ou d’électronique de faible consommation (par exemple asynchrone) constituent des pistes évidentes d’amélioration. / The reduction of power consumption in communication networks has become a key issue for both the Internet Service Providers (ISP) and the research community. Ac- cording to different studies, the power consumption of Information and Communication Technologies (ICT) varies from 2% to 10% of the worldwide power consumption [1, 2]. Moreover, the expected trends for the future predict a notably increase of the ICT power consumption, doubling its value by 2020 [2] and growing to around 30% of the worldwide electricity demand by 2030 according to business-as-usual evaluation scenarios [15]. It is therefore not surprising that researchers, manufacturers and network providers are spending significant efforts to reduce the power consumption of ICT systems from dif- ferent angles. To this extent, networking devices waste a considerable amount of power. In partic- ular, their power consumption has always been increased in the last years, coupled with the increase of the offered performance [16]. Actually, power consumption of network- ing devices scales with the installed capacity, rather than the current load [17]. Thus, for an ISP the network power consumption is practically constant, unrespectively to traffic fluctuations. However, actual traffic is subject to strong day/night oscillations [3]. Thus, many devices are underutilized, especially during off-peak hours when traffic is low. This represents a clear opportunity for saving energy, since many resources (i.e., routers and links) are powered on without being fully utilized. In this context, resource consolidation is a known paradigm for the reduction of the power consumption. It consists in having a carefully selected subset of network devices entering a low power state, and use the rest to transport the required amountof traffic.

Hearing conservation specialists are aware of the need for early identification and diagnosis of impaired hearing. This awareness of need has led to the development of several identification methods in the United States. Public awareness programs designed to inform laymen and professionals of the danger signals of infant hearing impairment currently are coming into focus, both as a separate entity and as part of total identification procedures. Current public awareness programs regarding infant hearing loss were surveyed in the present study and recommendations on a model awareness program of this type were obtained. Fifty-one hearing conservation specialists participated in the survey. The data from questionnaire returns indicated existence of eighteen programs from among the total respondents. It also shoed strong support for dissemination of pertinent information of hearing loss to the professional and parent populations of the United States. The data further revealed that program direction and finance should primarily be through state health departments with federal governmental assistance.

3D-aware image synthesis has attracted increasing interest as it models the 3D nature of our real world. However, performing realistic object-level editing of the generated images in the multi-object scenario still remains a challenge. Recently, a 2D GAN termed BlobGAN has demonstrated great multi-object editing capabilities on real-world indoor scene datasets. In this work, we propose BlobGAN-3D, which is a 3D-aware improvement of the original 2D BlobGAN. We enable explicit camera pose control while maintaining the disentanglement for individual objects in the scene by extending the 2D blobs into 3D blobs. We keep the object-level editing capabilities of BlobGAN and in addition allow flexible control over the 3D location of the objects in the scene. We test our method on real-world indoor datasets and show that our method can achieve comparable image quality compared to the 2D BlobGAN and other 3D-aware GAN baselines while being the first to enable camera pose control and object-level editing in the challenging multi-object real-world scenarios.

The integration of deep learning models and classical techniques in robotics is constantly creating solutions to problems once thought out of reach. The issues arising in most models that work involve the gap between experimentation and reality, with a need for strategies that assess the risk involved with different models when applied in real-world and safety-critical situations. This work proposes the use of Bayesian approximations of uncertainty from deep learning in a robot planner, showing that this produces more cautious actions in safety-critical scenarios. The case study investigated is motivated by a setup where an aerial robot acts as a "scout'' for a ground robot when the below area is unknown or dangerous, with applications in space exploration, military, or search-and-rescue. Images taken from the aerial view are used to provide a less obstructed map to guide the navigation of the robot on the ground. Experiments are conducted using a deep learning semantic image segmentation, followed by a path planner based on the resulting cost map, to provide an empirical analysis of the proposed method. The method is analyzed to assess the impact of variations in the uncertainty extraction, as well as the absence of an uncertainty metric, on the overall system with the use of a defined factor which measures surprise to the planner. The analysis is performed on multiple datasets, showing a similar trend of lower surprise when uncertainty information is incorporated in the planning, given threshold values of the hyperparameters in the uncertainty extraction have been met. / Master of Science / Deep learning (DL) is the phrase used to refer to the use of large hierarchical structures, often called neural networks, to approximate semantic information from data input of various forms. DL has shown superior performance at many tasks, such as several forms of image understanding, often referred to as computer vision problems. Deep learning techniques are trained using large amounts of data to map input data to output interpretation. The method should then perform correct input-output mappings on new data, different from the data it was trained on. Robots often carry various sensors from which it is possible to make interpretations about the environment. Inputs from a sensor can be high dimensional, such as pixels given by a camera, and processing these inputs can be quite tedious and inefficient given a human interpreter. Deep learning has recently been adopted by roboticists as a means of automatically interpreting and representing sensor inputs, like images. The issue that arises with the traditional use of deep learning is twofold: it forces an interpretation of the inputs even when an interpretation is not applicable, and it does not provide a measure of certainty with its outputs. Many techniques have been developed to address this issue with deep learning. These techniques aim to produce a measure of uncertainty associated with DL outputs, such that even when an incorrect or inapplicable output is produced, it is accompanied with a high level of uncertainty. To explore the efficacy and applicability of these uncertainty extraction techniques, this thesis looks at their use as applied to part of a robot planning system. Specifically, the input to the robot planner is an overhead image taken by an unmanned aerial vehicle (UAV) and the output is a path from a set start and goal position to be taken by an unmanned ground vehicle (UGV) below. The image is passed through a deep learning portion of the system that performs what is called semantic segmentation, mapping each pixel to a meaningful class, on the image. Based on the segmentation, each pixel is given a cost proportionate to the perceived level of safety associated with that class. A cost map is thus formed on the entire image, from which traditional robotics techniques are used to plan a path from start to goal. A comparison is performed between the risk-neutral case which uses the conventional DL method and the risk-aware case which uses uncertainty information accompanying the modified DL technique. The overall effects on the robot system are envisioned by observing a metric called the surprise factor, where a high surprise factor signifies a poor prediction of the actual cost associated with a path. The risk-neutral case is shown to have a higher surprise factor than the proposed risk-aware setup, both on average and in safety-critical case studies.