目前生物模擬大多採用物理計算來模擬生物的行為，相較真實生物而言，易讓人產生不真實之感，若透過人工事先建立生物模型之動畫，通常需大量人力來依照經驗設計，因此動畫品質易受人的影響，亦會受到動畫資料庫的限制，此外一般生物模擬僅有單方面的呈現，使得模擬過程過於單調。有鑑於此，我們提出一套使用視覺學習機制來學習真實生物的行為，使模型能學習影片中之生物行為以產生動畫，此外我們定義了數種互動手勢，透過偵測人的手勢觸發互動事件，並能讓人與虛擬世界互動。本論文中，我們以生物影像進行骨架作為輸入，對生物資訊進行分析與學習，來讓模型根據學習資訊產生一個不受限影片長度限制的生物模擬動畫，透過不同路徑與動作的安排，游動的呈現不會受限於取樣的資料量，亦不需事先建立動畫資料庫。此外我們使用Kinect設備來偵測人的骨架動作，以身體作為控制器來進行互動，無需昂貴的設備，亦不需複雜的設定，就能偵測人的手勢，即時產生互動之模擬動畫。

In this thesis, we presented a visual learning mechanism that is used to learn the realistic motion from real fish to generate 3D animation. Usually, biological motion is simulated by physical simulation, the results of motion do not appear realistically. Furthermore, artificial 3D animation which is built beforehand needs a huge amount of labors to design based on one’s experience. In that case, the quality of animation is susceptible to one’s experience and the quantity of motion is limited to the database. Therefore, we proposed an intelligent 3D fish animation system that uses a visual learning mechanism to learn behavior from realistic creature and an interaction mechanism to allow one to interact with the virtual fish. Two fish video sequences data from front and top views are used as inputs at the same time to derive the fish skeleton sequence. The proposed learning mechanism is used to analyze fish motion and create learning data. Fish behaviors are generated by arranging several paths according to learning data, so that the motion of virtual fish will not be limited by database. Furthermore, Kinect device is used to detect one’s gesture to obtain interaction events. Five interaction events --appearing, waving, feeding, scaring and stirring are defined to allow users to interact with the virtual 3D fish.

論文審定書 i
誌 謝 ii
摘 要 iii
Abstract iv
目 錄 v
圖目錄 vii
表目錄 ix
壹、 簡介 1
一. 論文概述 2
二. 論文貢獻 2
三. 論文架構 4
貳、 文獻探討 5
參、 研究方法 8
一. 偵測與辨識(Detection and Identification) 10
二. 事件偵測(Event Detection) 12
1、 偵測靠近事件 13
2、 偵測揮手事件 13
3、 偵測餵食事件 14
4、 偵測驚嚇事件 14
5、 偵測擾動事件 14
三. 學習機制(Learning Mechanism) 15
1、 資料分析(Data Analysis) 16
2、 建立事件學習對應表(Event Learning Table) 16
3、 建立事件狀態轉換機率表(State Transition Table) 17
四. 產生動畫 21
1、 事件決策 22
2、 路徑規劃 23
肆、 系統實作 25
一. 操作介面及環境 26
二. 實作詳細內容 28
伍、 結論 43
參考文獻 44

[1] T. C. Wu, “The Animation of Multi-Legged Creatures Using Inverse Kinematics,” Matser Thesis, Department of Computer Science and Engineering, Tatung University, 2012.
[2] D. Baraff, “Physically Based Modeling: Rigid Body Simulation,” ACM SIGGRAPH Course Notes, vol. 2, no. 1, 2001.
[3] R. Barzel, “Faking Dynamics of Ropes and Springs,” IEEE Computer Graphics and Applications, vol. 17, no. 3, pp. 31-39, 1997.
[4] X. Y. Tu, “Artificial Animals for Computer Animation Biomechanics, Locomotion, Perception, and Behavior,” Ph.D Thesis, Department of Computer Science, Toronto University, 1996.
[5] C. M. Kim, M. W. Shin, S. M. Jeong, I. C. Kang, E. S. Kim, and B. K. Kim, “Real-time Motion Simulation of Artificial Fish for Virtual Marine World,” IEEE International Conference on Advanced Language Processing and Web Information Technology, pp. 253-260, 2007.
[6] T. Masaki, T. Yamaguchi, and Y. Kitamura, “An Interactive Digital Fishtank Based on Live Video Images,” Advanced Multimedia Content Processing, pp. 386-396, 1998.
[7] A. Schödl, R. Szeliski, D. H. Salesin, and I. Essa, “Video Textures,” ACM Annual Conference on Computer Graphics and Interactive Techniques, pp. 489-498, 2000.
[8] T. Y. Lee and H. Y. Chen, “The Design and Implementation of Realistic Fish Animation Play System Using Video Sequence,” Journal of Information Science and Engineering, vol. 19, no. 6, pp. 1071-1085, 2003.
[9] C. C. Li and Y. Y. Chen, “Human Posture Recognition by Simple Rules,“ IEEE International Conference on Systems, Man and Cybernetics, pp. 3237-3240, 2006.
[10] C. Castiello, T. D'Orazio, A. M. Fanelli, P. Spagnolo, and M. A. Torsello, “A Model-free Approach for Posture Classification,“ IEEE Conference on Advanced Video and Signal Based Surveillance, pp. 276-281, 2005.
[11] A. Menache, Understanding Motion Capture for Computer Animation, 2 ed., Elsevier, 2011.
[12] N. Kinerd, “Motion Capture Study of Human Movement Recognition,” Master Thesis, Department of Computer Science, Clemson University, 2012.
[13] H. Harms, O. Amft, and G. Tröster, “Estimating Posture-Recognition Performance in Sensing Garments Using Geometric Wrinkle Modeling,” IEEE Transactions on Information Technology in Biomedicine, vol. 14, no. 6, pp. 1436-1445, 2010.
[14] D. U. Jeong, S. J. Kim, and W. Y. Chung, “Classification of Posture and Movement Using A 3-axis Accelerometer,“ IEEE International Conference on Convergence Information Technology, pp. 837-844, 2007.
[15] Microsoft, “Kinect for Windows SDK,” accessed 2015/02/03, available from http://msdn.microsoft.com/en-us/library/hh855347.aspx.
[16] Z. Y. Zhang, “Microsoft Kinect Sensor and Its Effect,” IEEE MultiMedia, vol. 19, no. 2, pp. 4-10, 2012.
[17] S. F. Huang, “Human's Postures Recognition by Using Kinect,” Master Thesis, Department of Electrical Engineering, National Central University, 2011.
[18] T. H. Li, S. H. Liu, J. S. Liu, and P. J. Chan, “Realistic Baseball - Kinect Motion Identification,“ International Conference on Digital Content, 2011.
[19] N. L. Hakim, “An Interactive 3D Facebook Visualization Tool Using Kinect,” Master Thesis, Department of Computer Science and Information Engineering, National Central University, 2013.
[20] S. C. Yang, “Human Action Recognition Using Kinect Camera,” Master Thesis, Departmeent of Information Engineering, I-Shou University, 2013.
[21] Paul Viola and M. J. Jones, “Robust Real-time Face Detection,” International Journal of Computer Vision, vol. 57, no. 2, pp. 137-154, 2004.
[22] W. C. Hsieh, “3D Fish Animation Using Video-based Visual Behavior Learning Mechanism,” Master Thesis, Department of Computer Science and Engineering, National Sun Yat-sen University, 2013.
[23] OpenCV, “Face Recognition with OpenCV,” accessed 2015/02/03, available from http://docs.opencv.org/trunk/modules/contrib/doc/facerec/facerec_tutorial.html.