在RoboCup中型機器人足球競賽中，機器人獲得其周圍環境資訊主要是透過裝置在自身頂部的全方位影像系統，為了應付競賽過程中複雜的環境變化，機器人影像的處理能力須達到即時性、穩定性、準確性之外更要注重良好的強健性，才能夠快速的做出正確的應對措施。因此，本論文提出以下三項視覺處理的方法以建構出整個中型足球機器人的視覺系統：(1)色彩模型的建立：針對競賽中多變的視覺環境，提出一套色彩模型建立的方法，使機器人能夠迅速的辨別出球場上特定的目標物體資訊，且抑制場地光源與其他環境的干擾、降低視覺系統的運算量。(2)目標物體的偵測：針對全方位影像資訊扭曲失真的特性，本論文提出極座標搜尋法的目標物搜尋方式，再搭配跳躍搜尋法以跳躍取樣的方式，能夠有效且迅速的獲得影像資訊中各個顏色的區塊資訊，透過對顏色區塊的分析來獲得各個目標物座標資訊，且能夠辨別球門與柱子等在機器視覺上容易混淆的目標物資訊。(3)自我定位的計算：本論文提出新的影像定位演算技術，僅需要利用影像資訊中三個任意目標物與影像中心的相對關係資訊，即可立即推算出機器人位於球場上絕對的座標位置，整個定位計算過程無需使用機器人與目標物之間的距離資訊。最後，本論文將視覺系統整合至自行研製完成的中型足球機器人內，透過機器人在實際球場上的移動，證明視覺系統其效能符合即時、穩定、準確、與強健等目標。 In this thesis, three design methods are proposed to construct the vision system of soccer robot for middle-size league of RoboCup: (1) Color model construction: A clustering-based method is proposed to effectively build a color model. It can let the vision system quickly identify objects on the field, inhibit the light interference in a changeful environment, and reduce the computation. (2) Objects detection: A polar coordinate scan method combined with a jump search method is proposed to detect objects in the captured image from an omni-directional mirror and determine their coordinates fast and effectively. The ball, goals, and corner cylinders in the captured image can be recognized by the proposed method. (3) Self-localization: A self-localization method based on three coordinates of targets extracted from the image data is proposed. It can determine the global coordinate of the robot quickly. In the practical experiments, a soccer robot with an omni-directional vision system and a four-wheeled omni-directional movement mechanism is constructed. The experiment results illustrate that the constructed vision system by the three proposed methods with the features of real-time, stable, accurate, and robust.
