本論文探討視覺式感測器輔助機器人巡航的議題，主要任務是輔助機器人進行自我定位與建立地圖，並且應用在全球定位系統(GPS)無法順利運作的環境之中。當機器人在環境中巡航時，視覺式感測器可以提供機器人狀態估測與建立環境地圖所需的量測訊息。本論文將使用雙眼攝影機建立雙眼視覺，透過非標準立體幾何(non-SSG)架構即時計算影像特徵點影像的深度，初始化地標點的三維座標。狀態估測器方面，將使用擴張型卡爾曼濾波器(EKF)，遞迴地預測與估測機器人與環境靜態物件的狀態。當地圖過大時會使擴張型卡爾曼濾波器運算量大幅上升而使運算速度無法達到即時性運算。將針對地圖過大時造成運算上無法達到即時性的問題，使用區域地圖方式，使EKF估測器的狀態向量維持在有限維度內，以降低電腦計算負擔。本研究將在PC-based控制器內建立所需的發展環境，以Visual C++程式語言整合視覺感測器、影像處理、與狀態估測器。整合的系統將可達到即時性應用於執行機器人於大環境下的同時定位與建圖之任務。 This thesis presents a vision-assisted robot navigation system. The major objective of the system is to assist the robot implementing the tasks of localization and mapping in the environment where the global positioning system (GPS) is denied. The visual sensor provided measurement data for estimating the robot state and building the environment map. The position of the landmarks was initialized using the non-standard stereo geometry method (non-SSG). The states of robot and static objects were recursively predicted and estimated using the extended Kalman filter. When the range of the environment map was too large, the computation time increased dramatically. Real-time implementation of robot visual navigation became an impossible task. To improve the problem, the concept of local map was proposed in this study. The sizes of the state and covariance vectors were limited in order to reduce the computation time. The software program of the robot navigation system was developed in a PC-based controller using Microsoft Visual Studio C++. The navigation system integrated the sensor inputs, image processing, and state estimation. The resultant system was used to perform the tasks of simultaneous localization and mapping (SLAM) for large environments.
