自主式移動機器人必須具備認知周遭環境、行動規劃及導航的能力,要具備此種能力,則機器人必須能夠在未知環境中同時定位及建構地圖。本論文使用單眼視覺以及里程計,進行移動機器人同步定位與建立地圖的相關研究,並以實驗驗證成效。 首先處理影像以擷取環境垂直方向物的邊緣特徵,並定義相關座標系,使三維環境降為二維,簡化運算時間,接著運用里程計計算機器人移動軌跡,建立機器人運動模型,本論文採用擴展卡爾曼濾波器消除感測器累計誤差及雜訊影響,對機器人狀態與特徵狀態進行估測以及更新,藉由環境中同一特徵於連續影像畫面中的匹配,以達到定位的目的,此外,本研究以雷射感測器同時繪製未知環境的地圖。
An autonomous mobile robot must possess capabilities of cognition, motion planning and navigation. Consequently, a robot has to localize its position and build map simultaneously in unknown environments. In this thesis, we study SLAM (simultaneously localization and mapping) of mobile robot employing monocular vision and odometer. Moreover, experiments were applied to evaluate the performance of the proposed mechanism. First the vision picture was processed to derive the objects’ vertical edge landmark. The relative coordinates were than defined and reduced from three-dimension to two-dimension to simply computation. Consequently, the robot’s trajectories derived using the data of odometer were applied to build the robot’s dynamic model. In this study, the extended Kalman filter was employed to eliminate the accumulated error from odometer measurement and noise. The states of mobile robot and landmarks were estimated and updated continuously. The localization of mobile robot was achieved through the matching of landmarks. Moreover, laser range finder was utilized to draw the map of unknown environment.