Human gestures, with the spatial-temporal variability, are difficult to be recognized by a generic model or classifier that are applicable for everyone. To address the problem, in this thesis, personalized dynamic gesture recognition approaches are proposed. Specifically, based on Dynamic Time Warping(DTW), a novel concept of Subject Relation Network is introduced to describe the similarity of subjects in performing dynamic gestures, which offers a brand new view for gesture recognition. By clustering or arranging training subjects based on the network, two personalization algorithms are proposed respectively for generative models and discriminative models. Moreover, three basic recognition methods, DTW-based template matching, Hidden Markov Model(HMM) and Fisher Vector combining classification, are compared and integrated into the proposed personalized gesture recognition. The proposed approaches are evaluated on a challenging dynamic hand gesture recognition dataset DHG14/28, which contains the depth images and skeleton coordinates returned by the Intel RealSense depth camera. Experimental results show that the proposed personalized algorithms can significantly improve the performance of basic generative&discriminative models and achieve the state-of-the-art accuracy of 86.2%. / Människliga gester, med spatiala/temporala variationer, är svåra att känna igen med en generisk modell eller klassificeringsmetod. För att komma till rätta med problemet, föreslås personifierade, dynamiska gest igenkänningssätt baserade på Dynamisk Time Warping (DTW) och ett nytt koncept: Subjekt-Relativt Nätverk för att beskriva likheter vid utförande av dynamiska gester, vilket ger en ny syn på gest igenkänning. Genom att klustra eller ordna träningssubjekt baserat på nätverket föreslås två personifieringsalgoritmer för generativa och diskriminerande modeller. Dessutom jämförs och integreras tre grundläggande igenkänningsmetoder, DTW-baserad mall-matchning, Hidden Markov Model (HMM) och Fisher Vector-klassificering i den föreslagna personifierade gestigenkännande ansatsen. De föreslagna tillvägagångssätten utvärderas på ett utmanande, dynamiskt handmanipulerings dataset DHG14/28, som innehåller djupbilderna och skelettkoordinaterna som returneras av Intels RealSense-djupkamera. Experimentella resultat visar att de föreslagna personifierade algoritmerna kan förbättra prestandan i jämfört medgrundläggande generativa och diskriminerande modeller och uppnå den högsta nivån på 86,2%.

As our society becomes more digitalized, the need for programmers is increasing. There are multiple schools world wide who have started implementing programming classes at early ages for their students. These classes are mostly taught using visual programming.Having the correct tools to teach programming in an introductory manner is of utmost importance, which visual programming can provide for the students.When novices start learning programming, the initial struggle for them is to master the aspects of syntax. Visual programming removes the aspect of syntax and focuses on teaching the students core concepts of programming. This is the problem that this project focuses on; The ability for novices to learn programming through visual blocks instead of syntax. The problem was approached with user testing of a game created, where the users solve puzzles using visual programming. Using the data from the user tests, evaluations and conclusions are drawn to whether visual programming is more beneficial for learning core programming concepts than syntax programming is.

Training is a vital link between idle or inadequately used software and software seen as a tool. Education can take many forms such as self exploration, CAI, manuals, and formal class room training. The introduction of a computer to a system interjects complex interactions between the computer user and his work environment. Research in this area has been performed by such disciplines as MIS, education, and psychology. A fundamental purpose of this research is to study the interaction of such user variables as cognitive style, demographics, and reading to variables from education and psychology such as advanced organizers, reinforcement, and document layout.

This thesis contributes to a better understanding of social media designs for more civil conversations online. I first demonstrated that people disengage from social media interactions when they encounter uncivil behavior from friends. To find alternative designs for social media that are more civil, I evaluated novel social interaction techniques. To do this, I designed a six-phase framework for prototyping social interactions called piggyback prototyping; and an algorithmic probe study methodology to include participants in the development of social curation algorithms. I built a piggyback prototype that modifies the civility on Facebook by highlighting positive posts in green and hiding impolite posts, and I deployed it as an algorithmic probe with 20 participants. I uncovered ways to improve the algorithm, and I found that participants responded most favorably to having civil posts highlighted. These findings open avenues for future research in designing pro-social platforms.

Uncertainty is inherent in many real-world settings; for example, in a combat situation, darkness may prevent a soldier from classifying approaching troops as friendly or hostile. In an environment plagued with uncertainty, decision-support systems, such as sensor-based networks, may make faulty assumptions about field conditions, especially when information is incomplete, or sensor operations are disrupted. Displaying the factors that contribute to uncertainty informs the decision-making process for a human operator, but at the expense of limited cognitive resources, such as attention, memory, and workload. This research applied principles of perceptual cognition to human-computer interface design to introduce uncertainty visualizations in an adaptive approach that improved the operator's decision-making process, without unduly burdening the operator's cognitive load. An adaptive approach to uncertainty visualization considers the cognitive burden of all visualizations, and reduces the visualizations according to relevancy as the user's cognitive load increases. Experiments were performed using 24 volunteer participants using a simulated environment that featured both intrinsic load, and characteristics of uncertainty. The experiments conclusively demonstrated that adaptive uncertainty visualization reduced the cognitive burden on the operator's attention, memory, and workload, resulting in increased accuracy rates, faster response times, and a higher degree of user satisfaction. This research adds to the body of knowledge regarding the use of uncertainty visualization in the context of cognitive load. Existing research has not identified techniques to support uncertainty visualization, without further burdening cognitive load. This research identified principles, such as goal-oriented visualization, and salience, which promote the use of uncertainty visualization for improved decision-making without increasing cognitive load. This research has extensive significance in fields where both uncertainty and cognitive load factors can reduce the effectiveness of decision-makers, such as sensor-based systems used in the military, or in first-responder situations.

Mobile computing devices are increasingly being utilized to support learning activities outside the traditional classroom environment. The text input capabilities of these devices represent a limiting factor for effective support of user-based interaction. The ability to perform continuous character selection and input to complete course exercises is becoming increasingly difficult as these devices become miniaturized to a point where traditional input and output methods are becoming less efficient for continuous text input. This study investigated the design and performance of a prototype mobile text entry keyboard (MobileType) based on characteristics of the linguistic frequency of character occurrence and increasing key size to minimize visual search time and distance during character selection. The study was designed to compare efficiency, effectiveness, and learning effects of the MobileType to the QWERTY keyboard layouts while performing fixed phrase and course exercise text entry tasks in two separate evaluation sessions. A custom software application was developed for a tablet device to display the two keyboard interfaces and capture text entry interaction and timing information. The results of this study indicated the QWERTY text entry interface performed faster in terms of efficiency, while the MobileType interface performed better in terms of effectiveness. In addition, there was an observable increase in the efficiency of the MobileType interface between the two task sessions. The results indicated that the MobileType interface was readily learnable relating to learning effect. Future research is recommended to establish if the performance of the MobileType interface could be increased with further participant familiarization after completing multiple sessions, which would validate the design of MobileType as a possible alternative to the QWERTY text entry interface for mobile devices.

The aim of this thesis is to locate current problems concerning the process of finding common ground between a creative producer and its respective client, furthermore it aims to propose a solution to this problem in the context of collaborative video editing. The paper starts off by exploring research related to the topic. After that, it establishes, through interviews with experts within fields of video editing and graphic design, what the currently existing problems are concerning communication within their line of work. As a solution to these problems, a collaborative software is proposed with the idea of bridging the understanding between the video editor and its client. The paper ends with some conclusions surrounding the current state of the topic and proposes a way forward for both practitioners and researchers.

published_or_final_version / abstract / Architecture / Doctoral / Doctor of Philosophy

Pas de résumé

Cette thèse porte sur 1'ínversion acoustique-articulatoire, c'est-à-dire la récupération des mouvements des articulateurs de la parole à partir du signal sonore. Nous présentons dans ce mémoire une évolution importante des méthodes de tabulation à codebooks utilisant une table de correspondants acoustique-articulatoire précalculée à l'aide d'un modèle de synthèse acoustique. En dehors de la méthode d'inversion proprement dite, nous présentons également l'introduction de deux types de contraintes : des contraintes phonétiques génériques, issues de l'analyse par des experts humains de l'invariance articulatoire des voyelles, et des contraintes visuelles, c'est-à-dire des contraintes obtenues automatiquement à partir de l'enregistrement et l'analyse d'images en stéréovision du locuteur.