Non-verbal natural interactive human intention estimation using time-varying fuzzy Markov models

機構典藏 > College of Engineering > Graduate Institute & Department of Electrical Engineering > Thesis > Item 987654321/88099

Please use this identifier to cite or link to this item: https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/88099

Title:	Non-verbal natural interactive human intention estimation using time-varying fuzzy Markov models
Other Titles:	時變模糊馬可夫模型於非語言自然互動之人類意向估測
Authors:	楊長恩;Yang, Chang-En
Contributors:	淡江大學電機工程學系碩士班劉寅春;Liu, Peter
Keywords:	非語言自然互動;模糊馬可夫模型;時變模糊馬可夫模型;意向辨識;意向機率模型;Fuzzy Markov model;time-varying system;intention inference engine
Date:	2012
Issue Date:	2013-04-13 12:00:41 (UTC+8)
Abstract:	本論文研究目的為建立非語言自然互動之意向估測模型，此模型將透過人類手部觸碰身體的位置座標估測蘊含的人類意向。基於人類的姿態動作改變初始的馬可夫意向機率，依此概念我們先後使用模糊馬可夫與時變模糊馬可夫機率模型，由於模糊馬可夫只適用於轉換與估測特定且具有規律的事件，其中只考慮現在與下一刻的狀態的機率。然而，人類的意向是一個時變的系統只能從人類的經驗或心理學統計資料分析概略的機率，為此我們提出時變模糊馬可夫模型估測人類的意向，此模組將透過人手與身體部位相互重疊時的座標作為輸入訊號，並經由模糊分析獲得相對應的模糊權重改變馬可夫機率，然後經由時變模組分析人手暫留於該部位的時間與整理環境的時間獲得時間機率權重，最後將模糊權重與時間機率權重相加總後調整人類的意向機率，如此一來將可獲知該部位所蘊含的非語言意向。依據上述方法我們獲得相同位置不同時間所產生的意向機率，此方法將使得人工智慧的決策機制更類似於人類的決策機制。於認知心理學中將人類的決策歸類區分為歸納推理與概率估算，而現今的人工智慧決策系統大部分使用歸納推理的方式獲得人類的資訊(例如：手勢辨識、智慧家庭控制等)。然而，我們所提出來的時變模糊馬可夫模型將運用歸納推理與概率估算人類的意向，如此一來將更類似人類的思考模式。若依此模組建構人類意向推論模型則必須要建構完整且合理的意向機率架構，所以本篇論文的意向辨識將與其它推論模型一樣侷限於初始意向機率模型或資料庫的大小。而本篇論文的意向辨識機率模型具備時變的特性所以更能模擬出類似人類的思考與決策模式，未來將再加以討論人類的自我學習機制，使意向辨識機率模型更加近似於人類的認知系統。 The estimation of human intention for robot decision mechanism is the ultimate goal of this thesis. The human decision mechanism most information to exist the non-verbal language in the human communication. If the human robot interaction via the human intention of non-verbal language estimation and analysis the information then the robot decision mechanism will be similarity the human thinking and reaction. Therefore, we propose time-varying fuzzy Markov model to estimate the human intention of meaning of posture. We will via MATLAB simulation the intention states of the hands touch body location purport the intention probability of behavior and emphasize the time-varying model into the intention inference system will be better then time-invariant inference system, because the human thinking will be varies with time. In this thesis, we establish a time-varying fuzzy Markov model to estimate human intention for natural non-verbal human robot interface. Based on human posture information, we change the probability between states to improve the accuracy of estimation of human intention. The advantages of the approach are three fold: i) non-verbal information is core of natural interaction; ii) time-varying probability improves estimation accuracy; and iii) fuzzy inference consider practical human experience.
Appears in Collections:	[Graduate Institute & Department of Electrical Engineering] Thesis

Files in This Item:

File	Size	Format
index.html	0Kb	HTML	166	View/Open

Loading...