【摘要】：隨著多媒體技術(shù)和虛擬現(xiàn)實技術(shù)的發(fā)展,人機環(huán)境中信息的輸出形式更加豐富,同時也使用戶所要面對的交互對象和交互內(nèi)容變得更加復(fù)雜,傳統(tǒng)的交互方式無法達(dá)到和諧、自然與人性化的交互要求。在軍事應(yīng)用領(lǐng)域,計算機輔助標(biāo)繪是是一個典型需求,亟需研究其他交互方式來提高標(biāo)繪交互自然性。本文融合手勢與語音識別技術(shù),對書空手勢指令進行定義和識別,構(gòu)建語音交互任務(wù)詞匯的狀態(tài)轉(zhuǎn)移矩陣,采用任務(wù)制導(dǎo)的方式整合不同通道的交互信息,提出了基于任務(wù)槽結(jié)構(gòu)的多通道整合模型,對交互任務(wù)和操作進行分析和設(shè)計,最后對交互任務(wù)進行綜合實驗。本文的主要工作和創(chuàng)新點有:一、提出了一種基于方向鏈碼的書空手勢識別算法,實現(xiàn)空間手勢識別。采用Leap Motion進行自定義的手勢識別和匹配,通過自定義手勢指令,對其自身有限的手勢識別指令進行擴充。為了消除手勢輸入過程中的不穩(wěn)定性而導(dǎo)致的噪聲干擾,對手勢軌跡進行分段處理,由分段的比重確定主要移動方向描述輸入手勢,根據(jù)手勢的相同分段對輸入手勢與模板手勢通過順序匹配算法進行匹配。二、在語音命令識別的基礎(chǔ)上,提出了基于命令轉(zhuǎn)移概率的語音任務(wù)組織方法,輔助語音命令識別和組織。根據(jù)語法規(guī)則和語義對交互任務(wù)語音詞匯進行分類,剔除語音交互任務(wù)中任務(wù)動作的生僻詞。通過場景語義上下文分析,確定當(dāng)前場景中的交互對象及交互任務(wù),采用馬爾可夫狀態(tài)轉(zhuǎn)移概率矩陣分析詞匯間的連接關(guān)系,排除異常輸入的關(guān)鍵詞,使系統(tǒng)能正確地理解用戶的語音交互意圖。三、提出了基于對象屬性的多通道任務(wù)槽結(jié)構(gòu)整合模型。對交互任務(wù)進行分析和設(shè)計,確定不同交互任務(wù)的任務(wù)槽的所需信息。用戶與傳感器進行元操作的交互,通過分層語義提取,將交互數(shù)據(jù)轉(zhuǎn)換為能夠被系統(tǒng)識別的任務(wù)所需的屬性信息。根據(jù)屬性類型的不同,將交互信息再填充到任務(wù)槽中相應(yīng)的模塊,構(gòu)成系統(tǒng)可識別的交互語義,從而識別整個交互任務(wù)并交由計算機執(zhí)行任務(wù),實現(xiàn)系統(tǒng)的交互功能。
[Abstract]:With the development of multimedia technology and virtual reality technology, the output form of information in man-machine environment is more abundant, meanwhile, the interaction object and content that users have to face become more complex, and the traditional interaction mode can not achieve harmony. Natural and human interaction requirements. In the field of military application, computer-aided plotting is a typical demand, so it is urgent to study other interactive methods to improve the natural nature of plotting interaction. This paper combines gesture and speech recognition technology, defines and recognizes the gesture instructions in the book space, constructs the state transition matrix of speech interactive task vocabulary, and integrates the interactive information of different channels by task-guided way. A multi-channel integration model based on task-slot structure is proposed to analyze and design interactive tasks and operations. Finally, a comprehensive experiment on interactive tasks is carried out. The main work and innovations of this paper are as follows: first, a novel algorithm of bookspace gesture recognition based on directional chain code is proposed to realize spatial gesture recognition. The self-defined gesture recognition and matching are carried out by Leap Motion, and the limited gesture recognition instruction is extended by using the self-defined gesture instruction. In order to eliminate the noise disturbance caused by the instability in gesture input, the gesture trajectory is segmented, and the main moving direction is determined to describe the input gesture. Input gesture and template gesture are matched by sequential matching algorithm according to the same segment of gesture. Secondly, on the basis of speech command recognition, a speech task organization method based on command transfer probability is proposed to assist speech command recognition and organization. According to the grammar rules and semantics, the phonetic vocabulary of interactive task is classified, and the unfamiliar words of task action in phonetic interaction task are eliminated. Through scene semantic context analysis, the interaction objects and interaction tasks in the current scene are determined. Markov state transition probability matrix is used to analyze the connection between words, and the keywords of abnormal input are excluded. So that the system can correctly understand the user's voice interaction intention. Thirdly, a multi-channel task slot structure integration model based on object attributes is proposed. Analyze and design interactive tasks to determine the information needed for different task slots. The user interacts with the sensor in meta-operation and transforms the interactive data into the attribute information needed by the task recognized by the system through hierarchical semantic extraction. According to the different attribute types, the interactive information is filled into the corresponding module in the task slot to form the interactive semantics which can be recognized by the system, so that the whole interactive task is recognized and the task is executed by the computer, and the interactive function of the system is realized.
【學(xué)位授予單位】：國防科學(xué)技術(shù)大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2014
【分類號】：TN912.3

【參考文獻】

相關(guān)期刊論文 前10條

1 曹磊;;一種三維空間手寫數(shù)字的融合識別方法[J];淮北師范大學(xué)學(xué)報(自然科學(xué)版);2013年04期

2 馬建平;潘俊卿;陳渤;;Android智能手機自適應(yīng)手勢識別方法[J];小型微型計算機系統(tǒng);2013年07期

3 張仲一;楊成;吳曉雨;;基于Kinect的隔空人手鍵盤輸入[J];中國傳媒大學(xué)學(xué)報(自然科學(xué)版);2013年03期

4 俞烈彬;孟凡文;;武器裝備系統(tǒng)中的人機交互新技術(shù)[J];電子世界;2013年12期

5 陳艷艷;陳正鳴;周小芹;;基于Kinect的手勢識別及在虛擬裝配技術(shù)中的應(yīng)用[J];電子設(shè)計工程;2013年10期

6 聶巖峰;田田;吳昊;;基于GIS指揮決策系統(tǒng)的多通道交互研究[J];計算機與現(xiàn)代化;2013年01期

7 賴英超;曾劍銘;沈海斌;;基于連筆消除的空間手寫字符識別方法[J];計算機工程;2012年19期

8 嚴(yán)軍;陳曉丹;沈海斌;;基于時頻融合特征的3D空間手寫識別[J];計算機工程;2012年18期

9 張毅;張爍;羅元;徐曉東;;基于Kinect深度圖像信息的手勢軌跡識別及應(yīng)用[J];計算機應(yīng)用研究;2012年09期

10 藍(lán)貴文;李景文;;基于ArcGIS Engine的可擴展地圖標(biāo)繪系統(tǒng)[J];桂林理工大學(xué)學(xué)報;2010年04期

，

本文編號：2230742