臺灣博碩士論文加值系統

本論文研究之目的在於透過倒傳遞神經網路的架構下，應用於桌球機器人之即時控制回擊上。實驗利用自行設計與製作之四軸機械臂，結合影像處理技術、立體視覺技術、類神經網路與機械手臂馬達控制，並於真實球桌大小之範圍，以實現桌球之回擊任務。
利用立體視覺量測得到桌球之三維座標後，本研究以類神經網路取代運動模型預測桌球回擊位置。本研究在PC架構下，配合影像擷取卡與CCD攝影機，以較低的成本，取代昂貴的三維影像擷取平台與複雜之回擊機構。由於桌球為三維空間之運動，故須建立空間座標，才能加以描述，於此使用左、右兩台攝影機來進行即時之影像處理。於實驗過程中須整合影像處理並計算三維空間座標，以獲得目標物資訊，藉由倒傳遞神經網路估測出球拍之回擊位置，最後驅動馬達令球拍到達擊球位置。經由實驗證實，類神經網路運用於估測目標物位置之可行性，並且在限定範圍與速度下，系統能完成回擊之任務。

The objective of the thesis is to construct a table tennis robot system that can be controlled under real time by using the back-propagation neural network architecture. In the experiment, a four-axis table tennis robot was designed and fabricated in house. The techniques of image processing, stereo vision, neural network, and motor motion control were integrated to achieve the hitting-back task by the table tennis robot.
The three-dimensional coordinates of the ball were first measured by using the stereo vision technique. Then, instead of using motion model, we employed the neural network to predict the hitting position of the racket. The thesis adopted the PC-based architecture together with video capture card and CCD cameras to lower the cost and avoid the expensive three-dimensional image acquisition platform. Due to the three-dimensional nature of the motion, a space coordinates measuring system must be established. Here we used two cameras to perform stereo vision and undertake the real-time image processing. In the experiment, the image processing and three-dimensional coordinates computation scheme were first performed to derive the target information, thereby the hitting position was estimated by the back-propagation neural network, and finally the racket was driven to the hitting position by commanding the motors. The feasibility of the above procedures was confirmed by experiments, that is, the system could achieve the hitting-back task under some restricted conditions.

第一章 緒論 1
1.1. 前言 1
1.2. 文獻回顧 3
1.3. 研究動機 7
1.4. 論文架構 8
第二章 理論背景 9
2.1. 影像處理 9
2.1.1 影像追蹤 9
2.1.2 色彩模型 10
2.1.3 濾光鏡 11
2.1.4 中值濾波器 12
2.2. 立體視覺系統 14
2.2.1 平面投影轉換 14
2.2.2 投影矩陣P求法 19
2.2.3 立體視覺重建世界座標 21
2.3. 類神經倒傳遞網路 23
2.3.1 類神經網路 23
2.3.2 倒傳遞演算法 24
第三章 實驗設備與架構 29
3.1. 硬體環境 29
3.1.1 電腦視覺系統 30
3.1.2 球拍運動控制系統 31
3.2. 軟體環境 36
3.3. 實驗架構 38
3.4. 實驗概述 39
3.4.1 目標物擷取方法 40
3.4.2 立體視覺建構 42
3.4.3 回擊位置估測 44
第四章 實驗及結果與討論 47
4.1. 影像擷取參數 47
4.2. 類神經參數 49
4.3. 實驗內容 56
4.4. 結果與討論 63
4.4.1 實驗討論與說明 63
4.4.2 系統效能與正確率 64
4.4.3 任務執行過程利弊分析 65
第五章 結論與未來展望 68
5.1. 結論 68
5.2. 未來展望 68
參考文獻 70
附錄A 立體視覺建構計算範例 74

[1]R. L. Andersson, “Aggressive trajectory generator for a robot ping-pong player”, Control Systems Magazine, IEEE Volume 9, pp. 15 – 21, 1989.
[2]R. L. Andersson, “Dynamic sensing in a ping-pong playing robot”, Robotics and Automation, IEEE Transactions onVolume 5, pp. 728 – 739, 1989.
[3]L. Acosta, J. J. Rodrigo, J. A. Mendez, G. N. Marichal and M. Sigut, “Ping-pong player prototype”, Robotics & Automation Magazine, IEEE. Volume. 10, pp. 44 – 52, 2003.
[4]K. P. Modi, F. Sahin and E. Saber, “An Application of Human Robot Interaction: Development of a Ping-Pong Playing Robotic Arm”, Systems, Man and Cybernetics, IEEE International Conference on Volume 2, pp. 1831 – 1836, 2005.
[5]F. Miyazaki, M. Takeuchi, M. Matsushima, T. Kusano and T. Hashimoto, “Realization of the Table Tennis Task based on Virtual Targets”, Robotics and Automation, IEEE International Conference on Volume 4, pp. 3844 - 3849 , 2002.
[6]M. Matsushima, T. Hashimoto, M. Takeuchi and F. Miyazaki, “Learning to the Robot Table Tennis Task—Ball Control & Rally with a Human—”, Robotics, IEEE Transactions on [see also Robotics and Automation, IEEE Transactions on] Volume 21, pp.767 – 771, 2005.
[7]M. Matsushima, T. Hashimoto, M. Takeuchi and F. Miyazaki, “A Learning Approach to Robotic Table Tennis”, IEEE Transactions on Robotics. Volume. 21. NO. 4, 2005.
[8]M. Takeuchi, F. Miyazaki, M. Matsushima, M. Kawatani and T. Hashimoto, “Dynamic Dexterity for the Performance of Wall-Bouncing Tasks”, Robotics and Automation, 2002. Proceedings. ICRA '02. IEEE International Conference on Volume 2, pp.1559 – 1564, 2002.
[9]G. Lei, “Recognition of Planar Objects in 3-D Space from Single Perspective Views Using Cross Ratio”, IEEE Trans. on Robotics and Automation, Volume. 6, No. 4, pp. 432-437, 1990.
[10]A. Fusiello, E. Trucco and A. Verri, “A Compact Algorithm for Rectification of Stereo Pairs”, Machine Vision and Applications, pp.16-22, 2002.
[11]I. H. Kim, D. E. Kim, Y. S. Cha, K. H. Lee and T. Y. Kuc, ”An Embodiment of Stereo Vision System for Mobile Robot for Real-time Measuring Distance and Object Tracking”, Control, Automation and Systems, ICCAS '07. International Conference on pp.1029 – 1033, 2007.
[12]Y. Z. Wang, D. L. Xie, C. X. Zhang, J. He and Y. Hu, ”Human Spine Profile Measurement by Means of Stereo Vision”, Machine Learning and Cybernetics, International Conference on13-16 Aug. 2006, pp.3944 – 3949, 2006.
[13]A. H. Rajpar, Q. Huang, Y. Pang, Y. Tian and K. Li, ”Location and Tracking of Robot End-effector Based onStereo Vision”, Robotics and Biomimetics, 2006. ROBIO '06. IEEE International Conference on17-20 Dec. 2006, pp.1140 – 1145, 2006.
[14]R. Tsai, “A Versatile Camera Calibration Technique for High-Accuracy 3D Machine Vision Metrology Using Off-the-Shelf TV Cameras and Lenses”, IEEE Trans. Robot. Automat., vol.3, pp. 323-344, 1987.
[15]R. Tsai’s camera calibration technique source code
URL:http://www-2.cs.cmu.edu/afs/cs.cmu.edu/user/rgw/www/TsaiCode.html
[16]Z. Zhang, ”A flexible new technique for camera calibration”, Pattern Analysis and Machine Intelligence, IEEE Transactions onVolume 22, pp.1330 – 1334,2000.
[17]D. Wenyi, L. Naiguang, Z. Jingcheng, G. Shuyi and Q. Xuguang, “A Stereo Vision System for Measurement of 3-D Object,” SPIE, Volume 3588, pp.248-252, 1998.
[18]R. A. Hicks and R. K. Perline, "Equiresolution Catadiopt ric Sensors", Applied Optics. 44 (29) (2005), pp. 6108-1114, 2005.
[19]C. J. Taylor and D. J. Kriegman, "Vision-Based Motion Planning and Explortion Algorithms for Mobile Rbots", IEEE Trans. On Robotics and Automation, 14 (3) (1998), pp.417-426, 1998.
[20]D. Murray and A. Basu, "Motion Tracking with an Active Camera", IEEE Transactions on Pattern Analysis and Machine Intelligence, 16 (5) (1994), pp. 449-459,1994.
[21]Y. K. Kim, H. C. Shin and J. Cho, “Real-Time Human Body Posture Estimation Using a Stereo Vision Embedded System”, Computational Advances in Multi-Sensor Adaptive Processing, 2007. CAMPSAP 2007. 2nd IEEE International Workshop on 12-14 Dec. 2007, pp.145 – 148, 2007.
[22]陳盈丞, 基於機器學習與視覺技術之桌上曲棍球機器人, 國立台灣海洋大學機械與機電工程研究所碩士論文, 2007.
[23]林翰屏, 基於機器學習與視覺技術之攻守型桌上曲棍球機器人, 國立台灣海洋大學機械與機電工程研究所碩士論文, 2008.
[24]何宜達, 視覺伺服技術於三維目標軌跡預測與攔截之應用, 國立成功大學航空太空工程學系暨研究所碩士論文, 2002.
[25]http://www.imshome.com/products/mdriveplus_overview.html
[26]繆紹綱編著, 數位影像處理, 普林斯頓國際股份有限公司, 2005.
[27]S. Haykin, Neural Networks: A Comprehensive Foundation (2nd Edition), 1998.
[28]N. Kazakova, M. Margala and N. G. Durdle, “Sobel edge detection processor for a real-time volume rendering system”, Circuits and Systems 2004 ISCAS 04 Proceedings of the 2004 International Symposium, Vol. 2, pp. 913-916, 2004.
[29]http://www.vision.caltech.edu/bouguetj/calib_doc/index.html
[30]繆紹綱編著, 數位影像處理活用Matlab, 全華科技, 2004.