博士学位論文

(1)

博士学位論文 Doctoral Dissertation

内容の要旨及び審査結果の要旨

Dissertation Abstracts and

Summaries of the Dissertation Review Results

第２４号

The Twenty-fourth Issue

平成２６年９月

September, 2014

The University of Aizu

(2)

はしがき

博士の学位を授与したので、学位規則（昭和２８年４月１日文部省令第９号）第８条の規定に基づき、その論文の内容の要旨及び論文審査の結果の要旨をここに公表する。

学位記番号に付した｢甲｣は学位規則第４条第１項（いわゆる課程博士）によるものであることを示す。

Preface

On granting the Doctoral Degree to the individuals mentioned below, abstracts of their theses and the theses review results are herewith publicly announced, in according to the provisions provided for in Article 8 of the Ruling of Degrees (Ministry Of Education Ordinance No.9, enacted on April 1, 1953)

The Chinese character, “甲”, at the beginning of the diploma number represents that an

individual has been granted the degree in accordance with the provisions provided for in

Paragraph 4-1 of the Ruling Of Degrees (what is called “Katei Hakase,” or the Doctoral

Degree granted by the University at which the grantee was enrolled.).

(3)

- 1 -

Abstract

Three research issues are discussed in this thesis including selection of wearable device, segmentation and recognition of continuous activity, and applications based on a finger-worn device.

A wearable device decides the types of detectable activities and wh ether it can be worn for long-term applications. In this thesis, we adopt a finger-worn device named Magic Ring (MR) for activity detection considering its high performance of activities detection from slight finger gestures to daily behaviors and low obstruction to users' daily lives.

Segmentation of activity sequence is a precondition in wearable computing. In this thesis, we propose an adaptive threshold-based method to segment short-duration, ambiguous, and individually different gestures. The method first obtains a proper initial threshold based on Bayes decision theory to ensure accurate initial segmentation, and then an adaptation mechanism is designed to automatically adjust the threshold to improve user-dependent accuracy.

Recognition is the resolution of wearable computing. We resolve recognition problem through two methods. One is feature-based recognition. We propose the concept of relative feature for quick and accurate recognition of finger gestures. The other one is model-based recognition.

We propose a probability model to classify gestures by a look-up-table approach for a small gesture set. Moreover, we propose a pre-classification HMM method for gesture recognition in a large gesture set. The method greatly reduces c omplexity of HMM method without any loss of accuracy.

Application is the final goal of wearable computing. In this thesis, we give two cases of MR-based applications. One case is appliance control. A gesture -based control system is proposed for one-for-all appliance control. The second case is ADLs (Activities of Daily Living) detection. We use only one MR to accurately detect ten ADLs. In addition, we propose a weighted sequence alignment approach to analyze activity sequences and their attributes to reflect realistic life aspects.

Summaries of the Dissertation Review Results

Using a finger-worn device for seamless interaction has attracted increasing attention consideration the flexibility of human fingers for rich semantic expression. However, there are still many unsolved problems in real-world applications. Under such a background, this thesis designed and developed a gesture recognition system based on a finger-worn wearable platform. The system can recognize gestures and daily activities and provide customized services to users through natural finger/hand gestures.

(6)

- 4 -

The thesis focuses on resolving some hard tasks, segmentation of continuous gestures, gesture and activity recognition, and gesture-based applications. The first task, segmentation, is to find out target gestures from a successive gesture sequence, which is a hard task to segment short-duration, ambiguous, and individually different gestures in daily living. To resolve the problem, the thesis studied the current methods of signal segmentation to propose an adaptive threshold-based segmentation method. The method showed high accuracy and low complexity in our experiments, which can save system energy for real application. The second task, recognition, is to know what the target gesture or activity is. To solve the problem, the thesis optimized the feature extraction, and proposed a pre-classification HMM method for large scale gesture recognition. The method greatly reduces complexity of HMM-based recognition method without any loss of accuracy. The third task, application, is to provide on-demand services to users. This thesis gave two cases, appliance control and ADLs (Activities of Daily Living) detection. The former service can replace remote controllers to control various appliances, and the later service can detect daily activities to identify abnormal behavior for healthcare application. The applications show the interest and significance of the research.

Currently, more and more researches focus on the human-centric service like voice-based communication, image-based control, brain-machine interface, and mobile-based applications. They are convenient but hard to overcome some problems like noise sensitive, privacy lost, hard to master, and handheld operation. Wearable computing is a hot research topic, but most of current researches lack flexibility and real-time utility ability. This research attempts to resolve these problems by using a mini wearable device to get on-demand services through natural gestures. That will greatly push forward development of wearable application and bring innovational changes to our lifestyles in forthcoming wearable era.

In summary, the thesis has been well organized and given clear written. The research progress and performance have been satisfied. During the doctoral dissertation review, the applicant gave clear, explicit and reasonable answers to the comments, and improved the research work and thesis according to their suggestions. The applicant has shown her ability to carry out independent and original research.

(7)

- 5 - Name

氏名

ZGONNIKOV, Arkady ズゴンニコフアルカディ The relevant degree

学位の種類

Doctoral degree (in Computer Science and Engineering) 博士（コンピュータ理工学）

Number of the diploma of the Doctoral Degree 学位記番号

甲CI博第43号

The Date of Conferment 学位授与日

September 19, 2014 平成26年9月19日 Requirements for Degree Conferment

学位授与の要件

Please refer to the article five of “University Regulation on University Degrees”

会津大学学位規程第５条該当 Dissertation Title

論文題目

Complex Dynamics of Human-Controlled Systems Caused by Uncertainty in Human Cognition

人間の不確実な認知特性に起因した人間－機械システムの複雑系ダイナミックス

Dissertation Review Committee Members 論文審査委員

University of Aizu,

Prof. LUBASHEVSKIY, I. (Chief Referee)

University of Aizu, Prof. KANEMOTO, S.

University of Aizu, Prof. MORI, K.

University of Aizu, Prof. ZHAO, Q.

会津大学教授イゴールルバシェフスキー（主査）

会津大学教授兼本茂会津大学教授森和好会津大学教授趙強福

(8)

- 6 -

Abstract

Understanding the fundamental properties of human control behavior is essential for identifying, investigating and preventing human errors in many critical procedures (e.g., aircraft landing) and various routines of everyday life (e.g., standing upright). Modern approaches to studying human control typically involve quantitative modeling and have been increasingly relying upon the notion of intermittent control. The latter is based on the intuition that humans do not react to small variations of the controlled system from the goal state. Despite gained success, modern theory of human control is still at its initial stage of development, and faces numerous challenges on its way to providing a comprehensive explanations of the dynamics of human-controlled systems. The present study faces two of such challenges.

First, there is a lack of cognitively plausible quantitative explanations of the control activation mechanism in humans. Contemporary models of intermittent human control employ the notion of threshold accompanied by noise, which allows such models to explain much of the experimentally observed dynamics. However, the present work argues that the mechanism of control activation in humans is heavily affected by cognitive uncertainty and therefore should be described not as deterministic, threshold-driven, but as stochastic, noise-driven. To support this key idea, a model of noise-driven control activation is developed. The model is confronted to the experimental data on virtual overdamped stick balancing, a novel task elaborated as an alternative to the standard experimental paradigms of studying human control. The model is shown to closely match the experimental data, which suggests that noise-driven control activation may be the key mechanism behind complex dynamics of human-controlled systems.

Second, there is no comprehensive understanding of the mechanisms leading to emergence of complex dynamics in the systems comprising many interacting operators. Previously it has been suggested that complex dynamics in the systems of interacting human operators can arise due to mismatch between the actions of individual operators. In this work this idea is elaborated further. Based on the novel idea of phase space extension, the action dynamical trap (ADT) model of operator's indifference to small variations of control strategy is developed. It is shown that such indifference can induce instability of the controlled system even in the case of single, isolated operator. Finally, it is demonstrated that ADT can cause unpredictable, chaotic dynamics even in the systems of just two elements. The results suggest that indifference to small variations of control strategy may be the key mechanism of emergence in the systems of interacting individuals.

Summaries of the Dissertation Review Results

The dissertation addresses several important issues in dynamics of human-controlled systems. First, the candidate develops a new approach to modeling control activation by human operators, which focuses on intrinsic stochasticity of this mechanism. Current approaches in the theory of human intermittent control employ the concept of threshold for modeling control activation. The candidate

(9)

- 7 -

shows experimentally that more complex mechanisms govern this process and suggests a

phenomenological model capturing dynamic competition between the status quo bias and the need to keep the controlled system close to the desired state. The model explains the experimental findings and provides grounds for further exploration of complex dynamics of human-controlled systems.

Second, the dissertation advances modern understanding of the dynamics of systems comprising many interacting individuals. The concept of action dynamical trap developed by the candidate is applied to examine emergent phenomena in the oscillatory chain with dynamical traps. The study demonstrates that indifference to small variations in action strategy can induce intricate cooperative dynamics even in small-scale multi-element systems.

The thesis review committee pointed out several directions in which the dissertation could be improved. Based on the committee’s feedback, the candidate revised the dissertation text, first, to enhance the presentation of the advantages of the proposed theoretical constructs and, second, to clarify the necessity of utilizing the non-conventional modeling techniques in describing dynamics of human-controlled systems. The final version of the dissertation was deemed to be of highest quality by the committee. The candidate as well has fulfilled all the formal requirements for the doctoral degree.

(10)

- 8 -

博士学位論文 Doctoral Dissertation

内容の要旨及び審査結果の要旨 Dissertation Abstracts

and

Summaries of the Dissertation Review Results

第24号

The Twenty-third Issue

平成２６年９月 September, 2014

発行会津大学

〒965-8580 福島県会津若松市一箕町鶴賀 TEL: 0242-37-2600

FAX: 0242-37-2526 THE UNIVERSITY OF AIZU Tsuruga, Ikki-machi Aizu-Wakamatsu City

Fukushima, 965-8580 Japan

博 士 学 位 論 文