利用靜態標的物全景影像軌跡資訊之室內定位系統

Abstract

現代機器人各種應用中,定位為一個重要的課題。機器人定位需要依賴著內部或是外部感測器的資訊，常用的感測器為視覺系統,例如攝影機。具有全方面視野範圍的全景攝影機與一般攝影機相較之下，更容易取得全面的環境資訊。 本論文提出一套使用單一視覺感測器的定位方法。靠著裝設在在移動機器人身上的全景攝影機所截取的連續影像，就可以讓機器人達到自我定位的功能。唯一需要的初始條件為標的物間的距離值。 首先，需要從全景影像中取得標的物相對於機器人的朝向角度。因此將連續的全景影像做幾何轉換以及取樣，再以時間順序組合成一張標的物的變化軌跡圖。利用此軌跡圖上的朝向角度值做為定位方法的輸入值，可估測機器人的自身位置。此定位方法的基本原理是根據機器人在不同位置時，所觀察到的3個標的物相對方向變化。 各種機器人移動路徑下的模擬結果證明此定位方法的可行性。而實際的實驗結果顯示，在較大的旋轉角度與移動距離時有著較好的定位結果，而小旋轉角度與小移動距離情形下會有著較大的誤差產生，原因是後者較容易受到攝影機未裝設在機器人旋轉中心上所造成的影響。 在已知標的物之間距離的貧乏條件下，本論文提出的視覺定位法只需要單一全景攝影機所截取的影像資訊，而不需要其他的感測器輔助。

Localization is an important issue in the robotic field generally. Robots localize themselves by their inner or outer sensor data. A common sensor is visual system, like cameras. In order to acquire the environment information completely, the omnidirectional camera is the best choice as the visual sensor for robot self-localization. A vision-based localization method with a single visual sensor is proposed in the thesis. Robots can localize themselves by a serious of omnidirectional images, and the only required is the distance between the landmarks. First, the landmark direction relative to robots has to be obtained from the omnidirectional images. The direction information will be the input of localization method. Therefore, the continuous omnidirectional images are transformed into panoramic images. And then a composition image called omnidirectional route panoramic map image are composed by the sampled data of panoramic images. Robots can successfully self-localize by the variation of directions of three landmarks at different positions. The correct simulation results of different moving routes are demonstrated. The experimental results show that the case of larger rotation angle and longer translation has higher accuracy. In contrast, the case of smaller rotation angle and shorter translation has lower accuracy. The latter result is gravely affected than the former one by the fact that the camera center is not at the rotation center of mobile robot. With the limited initial conditions, namely, the distances between landmarks, the localization method proposed in the thesis by using only the omnidirectional camera.