近年來無人載具(Unmanned vehicle)廣泛運用於急難搜救、地形空拍調查、巡邏監控等領域，同時工業型與家用型機器人載具也逐漸進入人類的日常生活中，擔任起輔助人類生產與生活的角色。當無人載具進入未知的環境後，僅能倚靠載具感測器基於環境中固定位置的特徵點的量測，進行載具自身位置的判定與環境模型的建立，唯單一影像僅包含二維資訊，不足以提供必要訊息重建環境與估算載具位置。

本論文研究目的是使用RGB相機並藉由電腦視覺影像處理技術進行相機之定位與建圖。首先定義世界座標系原點，並給與系統足夠已知特徵點，藉由座標系定義與相機校正之特性，推算相機初始世界坐標位置，並將影像儲存於資料庫中。隨著相機運動，已知特徵點逐漸減少，利用即時影像與資料庫影像中對應特徵點求得多張影像的Multi-view Geometry估算未知點的世界坐標，將估算結果代入計算，解構投影矩陣與相機外部參數，得到即時相機之世界坐標，並更新影像資料庫。

本論文僅使用單相機縮減系統架構，增加未來在不同載具應用上的彈性，減少搭載於載具之感測器數量，利用連續二維影像取得載具位置與環境資訊，減輕負重同時維持定位與建圖的精度與效率。

Unmanned Vehicles have been widely used in fields such as emergency search and rescue, aerial terrain mapping, reconnaissance and patrol tasks, also industrial robots and domestic robots have started taking part in human daily activities, playing roles in assisting production and domestic tasks. When an unmanned vehicle enters a total alien environment, information of the surrounding and the position of the vehicle are nowhere to be obtained, performing localization and building model of the surrounding relying solely on sensors on board the vehicle to measure fixed feature points in the environment. Yet single image contains only 2-dimensional information, not sufficient enough to determine vehicle’s position and build a model of the environment.

The purpose of this thesis is to perform localization and mapping based on a RGB camera and computer vision image processing techniques. First step is to define the origin of the world coordinate system, given enough feature points that contains sufficient scaling properties, through coordinate definition and properties of camera calibration, the initial position of the camera in the world coordinate system can be estimated, and the current image will be added to the image database. As the camera travels in the environment, the number of known feature points dropped below the necessary threshold for computation, by comparing the matching pairs of feature points in both the real-time image and history images and computing the Multi-view Geometry between these images we obtain coordinate information of the feature points existing in the environment, including estimated points into the computation, decomposing the projection matrix and camera’s extrinsic parameter we are able to estimate the position of the camera and update the image database.

This thesis uses only single camera to downsize the system architecture, retaining flexibility for future applications, using continuous 2-dimensional images to obtain camera position ( assume the same as the vehicle’s position) and information describing the environment, reducing payload while remain accurate and efficient localization and mapping operation.

[1] H.C. Longuet-Higgins, “Readings In Computer vision: Issues, problems, principles, and paradigms,” ch. A Computer Algorithm for Reconstruction a Scene from Reconstructing a Scene from Two Projections, pp.61-62, San Francisco, CA, USA: Mogran Kaufmann Publishers Inc., 1987.

[2] Faugeras, Three-dimensional computer Vision: A geometric Viewpoint. Cambridge, MA, USA: MIT Press, 1993.

[3] Z. Zhang, “A flexible new technique for camera calibration,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol.22, pp. 1330-1334, Nov. 2000.

[4] H. Durrant-Whyte and T. Bailey, “Simultaneous localization and mapping: part I,” Robotics Automation Magazine, IEEE, vol. 13, pp. 99-110, June 2006.

[5] T. Bailey and H. Durrant-Whyte, “Simultaneous localization and mapping (slam): part ii,” Robotics Automation Magazine, IEEE, vol. 13, pp. 108-117, Sept 2006.

[6] A. Davison, I. Reid, N. Molton , and O. Stasse, “Monoslam: Real-time Single Camera SLAM,” Pattern Analysis and Machine Intelligence, IEEE Transactions on, vol. 29, pp. 1052-1067, June 2007.

[7] 阮堯輝, 使用單一相機全向輪機器人之同時定位建圖與局部路徑規劃研究. PhD thesis, 中興大學, 台中市, 2008.

[8] 葉科廷, 單攝影機3D視覺定位. PhD thesis, 元智大學, 桃園縣, 2007.

[9] P. Henry, M. Krainin, E. Herbst, X. Ren, and D. Fox, “RGB-D Mapping: Using Kinect-style Depth Cameras for Dense 3D Modeling of Indoor Environment,” Int. J. Rob. Res., vol. 31, pp. 647-663, Apr. 2012.

[10] K. Muller, A. Smolic, M. Drose, P.Voigt, T. Wiegand, “3-D reconstruction of a dynamic environment with a fully calibrated background for traffic scenes,” Circuits and Systems for Video Technology, IEEE Transactions on, vol. 15, pp. 538-549, Apr. 2005.

[11] X. Zhang, B. Xian, B. Zhao, Y. Zhang, “ Autonomous Flight Control of a Nano Quadrotor Helicopter in a GPS-denied Environment Using on-board Vision,” to appear on Industrial Electronics, IEEE Transactions on, Apr. 2015

[12] 李其真, 基於資料庫影像與RGB-D 相機影像之同步定位與建圖. PhD thesis, 國立台灣科技大學, 台北市, 2013

[13] R. Hartley and A. Zisserman, Multiple View Geometry in Computer Vision. New York, NY, USA: Cambridge University Press, 2 ed.,2003.

[14] M. Fischler and R. Bolles, “Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,” Comm. Of the ACM, pp. 381-395, June 1981.

[15] Q. Luong and T. Viewille, “Canonical representations for the geometrics of multiple project views,” Computer vision and image understanding, vol. 64, no.2, pp. 193-229, 1996.