Gaze tracking technology is widely used in Human Computer Interaction applications such as in interfaces for assisting people with disabilities and for driver attention monitoring. However, commercially available gaze trackers are expensive and their performance deteriorates if the user is not positioned in front of the camera and facing it. Also, head motion or being far from the device degrades their accuracy. This thesis focuses on the development of real-time time appearance based gaze tracking algorithms using low cost devices, such as a webcam or Kinect. The proposed algorithms are developed by considering accuracy, robustness to head pose variation and the ability to generalise to different persons. In order to deal with head pose variation, we propose to estimate the head pose and then compensate for the appearance change and the bias to a gaze estimator that it introduces. Head pose is estimated by a novel method that utilizes tensor-based regressors at the leaf nodes of a random forest. For a baseline gaze estimator we use an SVM-based appearance-based regressor. For compensating the appearance variation introduced by the head pose, we use a geometric model, and for compensating for the bias we use a regression function that has been trained on a training set. Our methods are evaluated on publicly available datasets.

by Tsang Kwok Hang Elton. / Thesis submitted in: December 1998. / Thesis (M.Phil.)--Chinese University of Hong Kong, 1999. / Includes bibliographical references (leaves 78-83). / Abstract also in Chinese. / Chapter 1 --- Introduction --- p.1 / Chapter 2 --- Virtual-hand Recognition --- p.5 / Chapter 2.1 --- Hand model --- p.6 / Chapter 2.1.1 --- Hand structure --- p.6 / Chapter 2.1.2 --- Motions of the hand joints --- p.8 / Chapter 2.2 --- Hand-tracking technologies --- p.9 / Chapter 2.2.1 --- Glove-based tracking --- p.10 / Chapter 2.2.2 --- Image-based tracking --- p.12 / Chapter 2.3 --- Problems in virtual-hand recognition --- p.13 / Chapter 2.3.1 --- Hand complexity --- p.13 / Chapter 2.3.2 --- Human variations --- p.13 / Chapter 2.3.3 --- Immature hand-tracking technologies --- p.14 / Chapter 2.3.4 --- Time-varying signal --- p.14 / Chapter 2.3.5 --- Efficiency --- p.14 / Chapter 3 --- Previous Work --- p.16 / Chapter 3.1 --- Posture and gesture recognition algorithms --- p.16 / Chapter 3.1.1 --- Template Matching --- p.17 / Chapter 3.1.2 --- Neural networks --- p.18 / Chapter 3.1.3 --- Statistical classification --- p.20 / Chapter 3.1.4 --- Discontinuity matching --- p.21 / Chapter 3.1.5 --- Model-based analysis --- p.23 / Chapter 3.1.6 --- Hidden Markov Models --- p.23 / Chapter 3.2 --- Hand-input systems --- p.24 / Chapter 3.2.1 --- Gesture languages --- p.25 / Chapter 3.2.2 --- 3D modeling --- p.25 / Chapter 3.2.3 --- Medical visualization --- p.26 / Chapter 4 --- Posture Recognition --- p.28 / Chapter 4.1 --- Fuzzy concepts --- p.28 / Chapter 4.1.1 --- Degree of membership --- p.29 / Chapter 4.1.2 --- Certainty factor --- p.30 / Chapter 4.1.3 --- Evidence combination --- p.30 / Chapter 4.2 --- Fuzzy posture recognition system --- p.31 / Chapter 4.2.1 --- Objectives --- p.32 / Chapter 4.2.2 --- System overview --- p.32 / Chapter 4.2.3 --- Input parameters --- p.33 / Chapter 4.2.4 --- Posture database --- p.36 / Chapter 4.2.5 --- Classifier --- p.37 / Chapter 4.2.6 --- Identifier --- p.40 / Chapter 5 --- Performance Evaluation --- p.42 / Chapter 5.1 --- Experiments --- p.42 / Chapter 5.1.1 --- Accuracy analysis --- p.43 / Chapter 5.1.2 --- Efficiency analysis --- p.46 / Chapter 5.2 --- Discussion --- p.48 / Chapter 5.2.1 --- Strengths and weaknesses --- p.48 / Chapter 5.2.2 --- Summary --- p.50 / Chapter 6 --- Posture Database Editor --- p.51 / Chapter 6.1 --- System architecture --- p.51 / Chapter 6.1.1 --- Hardware configuration --- p.51 / Chapter 6.1.2 --- Software tools --- p.53 / Chapter 6.2 --- User interface --- p.54 / Chapter 6.2.1 --- Menu bar --- p.55 / Chapter 6.2.2 --- Working frame and data frame --- p.56 / Chapter 6.2.3 --- Control panel --- p.56 / Chapter 7 --- An Application: 3D Virtual World Modeler --- p.59 / Chapter 7.1 --- System Design --- p.60 / Chapter 7.2 --- Common operations --- p.62 / Chapter 7.3 --- Virtual VRML Worlds --- p.65 / Chapter 8 --- Conclusion --- p.70 / Chapter 8.1 --- Summaries on previous work --- p.70 / Chapter 8.2 --- Contributions --- p.73 / Chapter 9 --- Future Work --- p.75 / Bibliography --- p.78

Kwok, Tsz Ho. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2013. / Includes bibliographical references (leaves 161-175). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstracts also in Chinese.

Users of social networking services such as Facebook often want to manage the sharing of information and content with different groups of people based on their differing relationships. The growing popularity of such services has meant that users are increasingly faced with the copresence of different groups associated with different aspects of their lives, within their network of contacts. However, few users are utilising the group-based privacy controls provided to them by the SNS provider. In this thesis we examine the reasons behind the lack of use of group-based privacy controls, finding that it can be largely attributed to the significant burden associated with group configuration. We aim to overcome this burden by developing automated mechanisms to assist users with many aspects of group-based privacy control, including initial group configuration, labeling, adjustment and selection of groups for sharing privacy sensitive content. We use a mixed methods approach in order to understand: how automated mechanisms should be designed in order to support users with their privacy control, how well these mechanisms can be expected to work, what the limitations are, and how such mechanisms affect users’ experiences with social networking services and content sharing. Our results reveal the criteria that SNS users employ in order to configure their groups for privacy control and illustrate that off-the-shelf algorithms and techniques which are analogous to these criteria can be used to support users. We show that structural network clustering algorithms provide benefits for initial group configuration and that clustering threshold adjustments and detection of hubs and outliers with the network are necessary for group adjustment. We demonstrate that public profile data can be extracted from the network in order to help users to comprehend their groups, and that contextual information relating to context, contacts, and content can be used to make recommendations about which groups might be useful for disclosure in a given situation. We also show that all of these mechanisms can be used to significantly reduce the burden of privacy control and that users react positively to such features.

No description available.

Entrainment (sometimes called adaptation or alignment) is the tendency of human speakers to adapt to or imitate characteristics of their interlocutors' behavior. This work focuses on entrainment on acoustic-prosodic features. Acoustic-prosodic entrainment has been extensively studied but is not well understood. In particular, it is difficult to compare the results of different studies, since entrainment is usually measured in different ways, reflect- ing disparate conceptualizations of the phenomenon. In the first part of this thesis, we look for evidence of entrainment on a variety of acoustic-prosodic features according to various conceptualizations, and show that human speakers of both Standard American English and Mandarin Chinese entrain to each other globally and locally, in synchrony, and that this entrainment can be constant or convergent. We explore the relationship between entrainment and gender and show that entrainment on some acoustic-prosodic features is related to social behavior and dialogue coordination. In addition, we show that humans entrain in a novel domain, backchannel-inviting cues, and propose and test a novel hypothesis: that entrainment will be stronger in the case of an outlier feature value. In the second part of the thesis, we describe a method for flexibly and dynamically entraining a TTS voice to multiple acoustic-prosodic features of a user's input utterances, and show in an exploratory study that users prefer an entraining avatar to one that does not entrain, are more likely to ask its advice, and choose more positive adjectives to describe its voice. This work introduces a coherent view of entrainment in both familiar and novel domains. Our results add to the body of knowledge of entrainment in human-human conversations and propose new directions for making use of that knowledge to enhance human-computer interactions.

At least 360 million people worldwide have disabling hearing loss that frequently causes difficulties in day-to-day conversations. Hearing aids often fail to offer enough benefits and have low adoption rates. However, people with hearing loss find that speechreading can improve their understanding during conversation. Speechreading (often called lipreading) refers to using visual information about the movements of a speaker's lips, teeth, and tongue to help understand what they are saying. Speechreading is commonly used by people with all severities of hearing loss to understand speech, and people with typical hearing also speechread (albeit subconsciously) to help them understand others. However, speechreading is a skill that takes considerable practice to acquire. Publicly-funded speechreading classes are sometimes provided, and have been shown to improve speechreading acquisition. However, classes are only provided in a handful of countries around the world and students can only practice effectively when attending class. Existing tools have been designed to help improve speechreading acquisition, but are often not effective because they have not been designed within the context of contemporary speechreading lessons or practice. To address this, in this thesis I present a novel speechreading acquisition framework that can be used to design Speechreading Acquisition Tools (SATs) - a new type of technology to improve speechreading acquisition. I interviewed seven speechreading tutors and used thematic analysis to identify and organise the key elements of the framework. I evaluated the framework by using it to: 1) categorise every tutor-identified speechreading teaching technique, 2) critically evaluate existing Conversation Aids and SATs, and 3) design three new SATs. I then conducted a postal survey with 59 speechreading students to understand students' perspectives on speechreading, and how their thoughts could influence future SATs. To further evaluate the framework's effectiveness I then developed and evaluated two new SATs (PhonemeViz and MirrorMirror) designed using the framework. The findings from the evaluation of these two new SATs demonstrates that using the framework can help design effective tools to improve speechreading acquisition.

Learning and memory abilities decline with age; however collaborative learning with a familiar partner has been found to improve older adults’ performance on memory tasks and reduce these age-related differences. However it is unclear whether collaborating with a familiar partner is more beneficial to learning compared with collaborating with a stranger. Similarly, it is unclear whether older adults collaborate similarly with human and computer partners. The aim of this PhD thesis is to understand the role of collaborator identity on collaborative learning, and to investigate whether collaborative learning is as efficient and accurate with a range of learning partners. While collaborative learning is a socially-based memory task, the relationships between collaborative learning and social cognition have not yet been explored. The secondary aim of this thesis is to use experimental collaborative learning paradigms alongside standardised and experimental measures of social cognition to explore whether social cognition accounts for a significant amount of variance in collaborative learning performance with different partners. Two studies compare younger and older adults’ learning with familiar and unfamiliar partners on different collaborative learning paradigms. Two subsequent studies compare older adults’ learning on computerised versions of the collaborative learning tasks with partners they perceive to be humans or computers based on recordings of natural human or synthetic speech respectively. In all studies, measures of social cognition were used to assess whether social abilities affect learning outcomes with different partner types. When comparing older and younger adults’ results, familiarity had no effect on learning or immediate or delayed recall performance. Older adults initially took longer to complete the learning trials but performed with similar efficiency as younger adults by the final trials. Younger and older adults recalled collaboratively learned information with comparable accuracy after a delay of one hour, however after one week, older adults recalled the route less accurately than younger adults. Social cognition was not related to collaborative learning with familiar partners, but was related with unfamiliar partners, suggesting that those who are better able to take the perspective of another person may benefit during interactive learning. Social cognition was related to collaborative learning with perceived human partners but not perceived computer partners. This thesis offers a new perspective on the interplay between social and cognitive function in collaborative learning with different learning partners, and explores the differences between younger and older adults when learning collaboratively. The findings are discussed in relation to cognitive, social, and technological theories. On the whole, collaborative learning can result in older adults learning with similar speed and accuracy to younger adults; while familiarity does not improve learning outcomes, perceived human-ness does.

Current human-aware navigation approaches use a predominantly metric representation of the interaction which makes them susceptible to changes in the environment. In order to accomplish reliable navigation in ever-changing human populated environments, the presented work aims to abstract from the underlying metric representation by using Qualitative Spatial Relations (QSR), namely the Qualitative Trajectory Calculus (QTC), for Human-Robot Spatial Interaction (HRSI). So far, this form of representing HRSI has been used to analyse different types of interactions online. This work extends this representation to be able to classify the interaction type online using incrementally updated QTC state chains, create a belief about the state of the world, and transform this high-level descriptor into low-level movement commands. By using QSRs the system becomes invariant to change in the environment, which is essential for any form of long-term deployment of a robot, but most importantly also allows the transfer of knowledge between similar encounters in different environments to facilitate interaction learning. To create a robust qualitative representation of the interaction, the essence of the movement of the human in relation to the robot and vice-versa is encoded in two new variants of QTC especially designed for HRSI and evaluated in several user studies. To enable interaction learning and facilitate reasoning, they are employed in a probabilistic framework using Hidden Markov Models (HMMs) for online classiffication and evaluation of their appropriateness for the task of human-aware navigation. In order to create a system for an autonomous robot, a perception pipeline for the detection and tracking of humans in the vicinity of the robot is described which serves as an enabling technology to create incrementally updated QTC state chains in real-time using the robot's sensors. Using this framework, the abstraction and generalisability of the QTC based framework is tested by using data from a different study for the classiffication of automatically generated state chains which shows the benefits of using such a highlevel description language. The detriment of using qualitative states to encode interaction is the severe loss of information that would be necessary to generate behaviour from it. To overcome this issue, so-called Velocity Costmaps are introduced which restrict the sampling space of a reactive local planner to only allow the generation of trajectories that correspond to the desired QTC state. This results in a exible and agile behaviour I generation that is able to produce inherently safe paths. In order to classify the current interaction type online and predict the current state for action selection, the HMMs are evolved into a particle filter especially designed to work with QSRs of any kind. This online belief generation is the basis for a exible action selection process that is based on data acquired using Learning from Demonstration (LfD) to encode human judgement into the used model. Thereby, the generated behaviour is not only sociable but also legible and ensures a high experienced comfort as shown in the experiments conducted. LfD itself is a rather underused approach when it comes to human-aware navigation but is facilitated by the qualitative model and allows exploitation of expert knowledge for model generation. Hence, the presented work bridges the gap between the speed and exibility of a sampling based reactive approach by using the particle filter and fast action selection, and the legibility of deliberative planners by using high-level information based on expert knowledge about the unfolding of an interaction.

Yim-Pan Chui. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2001. / Includes bibliographical references (leaves 126-132). / Abstracts in English and Chinese. / Abstract --- p.ii / Acknowledgments --- p.v / Chapter 1 --- Introduction --- p.1 / Chapter 1.1 --- Introduction to Collaborative Virtual Environments --- p.1 / Chapter 1.2 --- Barriers to Resource Management and Optimization --- p.3 / Chapter 1.3 --- Thesis Contributions --- p.5 / Chapter 1.4 --- Application of this Research Work --- p.6 / Chapter 1.5 --- Thesis Organization --- p.6 / Chapter 2 --- Resource Optimization - Intelligent Server Partitioning --- p.9 / Chapter 2.1 --- Introduction --- p.9 / Chapter 2.2 --- Server Partitioning --- p.13 / Chapter 2.2.1 --- Related Works --- p.15 / Chapter 2.2.2 --- Global Optimization Approaches --- p.17 / Chapter 2.3 --- Hybrid Genetic Algorithm Paradigm --- p.17 / Chapter 2.3.1 --- Drawbacks of traditional GA --- p.18 / Chapter 2.3.2 --- Problem Modeling --- p.19 / Chapter 2.3.3 --- Discussion --- p.24 / Chapter 2.4 --- Results --- p.25 / Chapter 2.5 --- Concluding Remarks --- p.28 / Chapter 3 --- Dynamic State Management - Dead Reckoning of Attitude --- p.32 / Chapter 3.1 --- Introduction to Dynamic State Management --- p.32 / Chapter 3.2 --- The Dead Reckoning Approach --- p.35 / Chapter 3.3 --- Attitude Dead Reckoning by Quaternion --- p.37 / Chapter 3.3.1 --- Modeling of the Paradigm --- p.38 / Chapter 3.3.2 --- Prediction Step --- p.39 / Chapter 3.3.3 --- Convergence Step --- p.40 / Chapter 3.3.4 --- Overall Algorithm --- p.46 / Chapter 3.4 --- Results --- p.47 / Chapter 3.5 --- Conclusion --- p.51 / Chapter 4 --- Polynomial Attitude Extrapolation --- p.52 / Chapter 4.1 --- Introduction --- p.52 / Chapter 4.2 --- Related Works on Kalman Filtering --- p.53 / Chapter 4.3 --- Historical Propagation of Quaternion --- p.54 / Chapter 4.3.1 --- Cumulative Extrapolation --- p.54 / Chapter 4.3.2 --- Method I. Vandemonde Approach --- p.55 / Chapter 4.3.3 --- Method II. Lagrangian Approach --- p.58 / Chapter 4.4 --- History-Based Attitude Management --- p.60 / Chapter 4.4.1 --- Multi-order Prediction --- p.60 / Chapter 4.4.2 --- Adaptive Attitude Convergence --- p.63 / Chapter 4.4.3 --- Overall Algorithm --- p.67 / Chapter 4.5 --- Results --- p.69 / Chapter 4.6 --- Conclusion --- p.77 / Chapter 5 --- Forward Difference Approach on State Estimation --- p.78 / Chapter 5.1 --- Introduction --- p.78 / Chapter 5.2 --- Positional Forward Differencing --- p.79 / Chapter 5.3 --- Forward Difference on Quaternion Space --- p.80 / Chapter 5.3.1 --- Attitude Forward Differencing --- p.83 / Chapter 5.3.2 --- Trajectory Blending --- p.84 / Chapter 5.4 --- State Estimation --- p.86 / Chapter 5.5 --- Computational Efficiency --- p.87 / Chapter 5.6 --- Results --- p.88 / Chapter 5.7 --- Conclusion --- p.96 / Chapter 6 --- Predictive Multibody Kinematics --- p.98 / Chapter 6.1 --- Introduction --- p.98 / Chapter 6.2 --- Dynamic Management of Multibody System --- p.100 / Chapter 6.2.1 --- Multibody Representation --- p.100 / Chapter 6.2.2 --- Paradigm Overview --- p.101 / Chapter 6.3 --- Motion Estimation by Joint Extrapolation --- p.102 / Chapter 6.3.1 --- Individual Joint Extrapolation --- p.102 / Chapter 6.3.2 --- Forward Propagation of Joint State --- p.104 / Chapter 6.3.3 --- Pose Correction --- p.107 / Chapter 6.4 --- Limitations on Predictive Articulated State Management --- p.108 / Chapter 6.5 --- Implementation and Results --- p.109 / Chapter 6.6 --- Conclusion --- p.112 / Chapter 7 --- Complete System Architecture --- p.113 / Chapter 7.1 --- Server Cluster Model --- p.113 / Chapter 7.1.1 --- Peer-Server Systems --- p.114 / Chapter 7.1.2 --- Server Hierarchies --- p.114 / Chapter 7.2 --- Multi-Level Resource Management --- p.115 / Chapter 7.3 --- Aggregation of State Updates --- p.116 / Chapter 7.4 --- Implementation Issues --- p.117 / Chapter 7.4.1 --- Medical Visualization --- p.117 / Chapter 7.4.2 --- Virtual Walkthrough Application --- p.118 / Chapter 7.5 --- Conclusion --- p.119 / Chapter 8 --- Conclusions and Future directions --- p.121 / Chapter 8.1 --- Conclusion --- p.121 / Chapter 8.2 --- Future Research Directions --- p.122 / Chapter A --- Quaternion Basis --- p.124 / Chapter A.1 --- Basic Quaternion Mathematics --- p.124 / Chapter A.2 --- The Exponential and Logarithmic Maps --- p.125 / Bibliography --- p.126