本論文主要實現一個於戶外動態環境下能夠自主行走、自動避障及導航的機器人系統。整體的系統架構使用筆記型電腦為主要的控制核心，採用深度學習、影像處理以及馬達控制等技術，由一台攝影機和手機作為路徑規劃與導引的依據。
戶外導航機器人採用輪型移動型態，另行設計一個機器人平台，置放所需的硬體設備。機器人由兩個主動輪控制方向行進，另一個被動輪保持車身穩定。基於Google Maps API自行撰寫手機APP，由手 機的GPS和電子羅盤感測器，規劃全域性的路線導航，並取得機器人即時的經緯度位置與方向角作為行進依據。障礙物距離偵測主要基於深度學習技術，由雙眼攝影機擷取的影像作為深度卷積神經網絡的訓練資料，其輸出結果是單張影像的視差圖，使用立體視覺幾何關係轉換距離，透過倒傳遞類神經網路針對實際距離再訓練，取得影像中每個像素的實際距離，即可讓機器人上的網路攝影機辨識障礙物距離。在辨識障礙物方面，由語意分割來分辨影像中道路與障礙物。使用模糊理論計算欲切除會造成誤判的岔路區域，最後綜合所有資訊來計算機器人行走的軌跡點，根據此軌跡，使用線性控制來控制馬達，使機器人能夠即時應變並根據規劃的路線行進，以完成自行避障前進的功能。
使用者能夠自行選擇目的地，機器人會依照APP規劃的全域路線與即時的路況，使戶外導航機器人有效率地避開障礙物且自動抵達目的地。

This thesis presents an outdoor navigation robot system that can automatically go forward, avoid obstacles and navigate. The laptop is the main controller, and a camera and a smartphone are used to plan path. The system combines advanced technologies such as deep learning, image processing and motor control.
The outdoor navigation robot adopts three-wheeled mobile robot. Two driving wheels are used to control the direction and the other passive wheel is used to keep the robot stability. The smartphone application utilizes the Google Maps API, the GPS and the electronic compass sensors get the global route planning. The current latitude, longitude position and direction of the robot are taken for navigation. Obstacle distance detection is mainly based on deep learning technology. The training data images are captured by the stereo camera, and output results are disparity of the single image. Then, the distances are converted by using the triangulation method of computer vision. The back propagation neural network retrains to obtain the actual distance of each pixel in the image. Therefore, the robot with the monocular camera could know the distance between obstacles and itself. Semantic segmentation is utilized to a to distinguish road and obstacles in the image. Fuzzy theory for calculating the area of the road which be cut is designed to avoid walking into the intersection.
The navigation trajectory of the robot is computed by all the information. According to the trajectory, the robot can immediately follow the planned route. Finally, the robot can walk along the planned path.
Users can select a destination with smartphone application. With the technologies of deep learning and image processing, the outdoor navigation robot can effectively avoid obstacles and arrive at the destination automatically.

[1]C. Caraffi, S. Cattani and P. Grisleri, "Off-road path and obstacle detection using decision networks and stereo vision," IEEE Transactions on Intelligent Transportation Systems, vol. 8, no. 4, pp. 607-618, Dec 2007.
[2] C. Siagian, C. K. Chang and L. Itti, "Mobile robot navigation system in outdoor pedestrian environment using vision-based road recognition," IEEE International Conference on Robotics and Automation, Karlsruhe, pp. 564-571, 2013.
[3] A. Cherubini, F. Spindler and F. Chaumette, "Autonomous visual navigation and laser-based moving obstacle avoidance," IEEE Transactions on Intelligent Transportation Systems, vol. 15, no. 5, pp. 2101-2110, Oct 2014.
[4] S. Duan et al., "Research on obstacle avoidance for mobile robot based on binocular stereo vision and infrared ranging," 2011 The 9th World Congress on Intelligent Control and Automation, Taipei, pp. 1024-1028, 2011.
[5] C. H. Chao et al., and T. H. S. Li, "Fuzzy target tracking and obstacle avoidance of mobile robots with a stereo vision system," International Journal of Fuzzy Systems, vol. 11, no. 3, pp. 183-191, 2009.
[6] Y. Liu et al., "Navigation research on outdoor miniature reconnaissance robot," 2016 IEEE International Conference on Mechatronics and Automation, Harbin, pp. 977-982, 2016.
[7] D. Gardeazabal, V. Ponomaryov, and I. Chairez, “Fuzzy control for obstacle avoiding in mobile robots using stereo vision algorithms,” IEEE 2011 Electrical Engineering Computing Science and Automatic Control, pp. 1–6, October 2011.
[8] A. Saxena, M. Sun and A. Y. Ng, "Make3D: Learning 3D scene structure from a single still image," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 31, no. 5, pp. 824-840, May 2009.
[9] D. Eigen, C. Puhrsch, and R. Fergus, "Depth map prediction from a single image using a multi-scale deep network," Neural Information Processing Systems, 2014.
[10] D. Eigen and R. Fergus, "Predicting depth, surface normals and semantic labels with a common multi-scale convolution al Architecture," 2015 IEEE International Conference on Computer Vision (ICCV), Santiago, pp. 2650-2658, 2015.
[11] J. Flynn, I. Neulander, J. Philbin and N. Snavely, "Deep stereo: learning to predict new views from the world's imagery," 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, pp. 5515-5524, 2016.
[12] F. Liu, Chunhua Shen and Guosheng Lin, "Deep convolutional neural fields for depth estimation from a single image," 2015 IEEE Conference on Computer Vision and Pattern Recognition, Boston, MA, pp. 5162-5170, 2015.
[13] D. Xu et al., “Multi-scale continuous CRFs as sequential deep networks for monocular depth estimation,” arXiv preprint, arXiv:1704.02157, 2017.
[14] J. Xie, R. Girshick, and A. Farhadi, "Deep3d: Fully automatic 2d-to-3d video conversion with deep convolutional neural networks," European Conference on Computer Vision, 2016.
[15] C. Godard, O. M. Aodha and G. J. Brostow, "Unsupervised monocular depth estimation with left-right consistency," 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, HI, pp. 6602-6611, 2017.
[16] K. He et al., "Deep residual learning for image recognition," 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, pp. 770-778, 2016.
[17] D.-A. Clevert, T. Unterthiner, and S. Hochreiter, "Fast and accurate deep network learning by exponential linear units (ELUs).", arXiv preprint arXiv:1511.07289, 2015.
[18] R. Hartley and A. Zisserman, Multiple View Geometry in Computer　Vision, Cambridge University Press, 2000.
[19] N. A. Campbell and R. A. Dunne, "On the pairing of the softmax activation and cross-entropy penalty functions and the derivation of the softmax activation function", Proc. 8th Aust. Conf. on the Neural Networks, Melbourne, 181–185, 1997.
[20] 林宜臻，基於深度學習之戶外導航機器人，碩士論文，中央大學，台北，2018年1月。
[21] A. Paszke, A. Chaurasia, S. Kim, and E. Culurciello, "Enet: A deep neural network architecture for real-time semantic segmentation," arXiv preprint arXiv:1606.02147, 2016.
[22] Android studio相關網站，https://developer.android.com/studio/，2018年6月。
[23] Google Maps API相關網站，https://code.google.com/apis/maps/，2018年6月。
[24] Van Brummelen, Glen Robert, Heavenly Mathematics: The Forgotten Art of Spherical Trigonometry, Princeton University Press. ISBN 9780691148922. 0691148929.
[25] 王文俊，認識 Fuzzy-第三版，全華科技圖書股份有限公司，2008 年 6 月。
[26] 王文中, 統計學與 Excel 資料分析之實習應用<第五版>, 博碩文化股份有限公司, 2004
[27] Zhou Wang, A. C. Bovik, H. R. Sheikh and E. P. Simoncelli, "Image quality assessment: from error visibility to structural similarity," IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600-612, April 2004.