王傑智Wang, Chieh-Chih臺灣大學:資訊工程學研究所陳俊維Chen, Chun-WeiChun-WeiChen2010-06-022018-07-052010-06-022018-07-052008U0001-2207200814161100http://ntur.lib.ntu.edu.tw//handle/246246/184925近年來，人與機器人互動成為機器人學及相關領域中熱烈討論的一個重要課題。由於人臉可能是人類最富有表現自己想法的部位，因此，觀察人臉成為了人與機器人互動中最重要的任務之一。過去二十年，主動外觀模型（Active Appearance Model）成功地建立了人臉模型並且能在影像中找到人臉細部特徵的位置，如眼睛、鼻子和嘴巴。主動外觀模型可成為分析與瞭解人類表情的重要工具之一。但現有主動外觀模型利用平面形狀模型來為立體的人臉建模，其中立體的資訊就被忽略了。因此，當像機拍攝到非正臉的人臉時，就有極高可能無法正確地找到這些特徵。在此篇論文中，我們提出三維立體主動外觀模型來克服此問題。我們所提出之三維立體主動外觀模型包含了一立體形狀模型（3D Shape Model）以及外觀模型（Appearance Model）。此模型於訓練階段時僅需要正臉的人臉資料即可。即使在有轉動姿勢的人臉的情況下，也可成功地定位人臉的細部特徵。經由20個人的資料庫所訓練並測試的實驗結果顯示出所提出的演算法的成功辨識率更可達80%。Perceiving human faces is one of the most important functions for human robot interaction. The active appearance model (AAM) is a statistical approach that models the shape and texture of a target object. According to a number of the existing works, AAM has a great success in modeling human faces. Unfortunately, the traditional AAM framework could fail when the face pose changes as only 2D information is used to model a 3D object. To overcome this limitation, we propose a 3D AAM framework in which a 3D shape model and an appearance model are used to model human faces. Instead of choosing a proper weighting constant to balance the contributions from appearance similarity and the constraint on consistent 2D shape with 3D shape in the existing work, our approach directly matches 2D visual faces with the 3D shape model. No balancing weighting between 2D shape and 3D shape is needed. In addition, only frontal faces are needed for training and non-frontal faces can be aligned successfully. The experimental results with 20 subjects demonstrate the effectiveness of the proposed approach.ABSTRACT iiIST OF FIGURES vIST OF TABLES viHAPTER 1. Introduction 1HAPTER 2. Related Works 3HAPTER 3. 2D Active Appearance Model 5.1. 2D Shape Model 5.2. Appearance Model 6.3. Pose Variations 7.4. Texture Transformation between Two Shapes 7.5. Fitting an AAM to an Image 8HAPTER 4. 3D Active Appearance Model 10.1. 3D Active Appearance Model 10.1.1. 3D Shape Model 10.1.2. Mapping from 3D Shape to 2D Shape 11.1.3. Texture Transformation between Two Shapes 11.2. Fitting an 3D AAM to an Image 12.2.1. Linearization of the Cost Function 13.2.2. Minimization of the Cost Function 13.2.3. Warp Jacobian 14HAPTER 5. Experimental Results 17.1. Data Collection 17.2. Software and Hardware 19.3. Experiments 19.4. Results 20.4.1. The Person-Specific Case 20.4.2. The Multi-Person Case 21.4.3. Glasses 22.4.4. 3D Pose Variations 22.5. Computational Complexity 22HAPTER 6. Conclusions and Future Works 33.1. Conclusions 33.2. Future Works 33IBLIOGRAPHY 34application/pdf3070547 bytesapplication/pdfen-US主動外觀模型立體主動外觀模型電腦視覺人機互動Active Appearance Model3D Active Appearance ModelComputer VisionHuman Robot Interactionthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/184925/1/ntu-97-R94922132-1.pdf