在室內環境使用特殊號誌辨識技術作自動車航行

Abstract

本論文提出以電腦視覺技術，辨識特殊號誌，完成一套適用於室內走廊環 境的自動車導航技術。我們選擇了數種走廊最常出現的物體當作特殊號誌 （如盆栽､門､垃圾桶､天花板電燈) ，及一經設計的轉彎標誌，對每一 個號誌，依需求賦予一個特殊意義（如轉彎或停止等) ，則一個有計畫的 航行可經由一組有順序排列的號誌來完成。在系統中，先以我們所提出之 模組學習技巧來抽取特殊號誌特徵，並儲存起來，在之後導航時再以模組 比對方式，完成號誌的辨識，而使自動車具有按特殊號誌之意義行動之能 力。直線導航方面，我們利用天花板電燈抽取走廊中心線，進而利用此直 線來進行直線航行。在轉彎處，我們結合以上兩種技術所得到的資訊以及 事先學習的轉彎資訊使得自動車具有盲目轉彎的能力。最後實際應用本論 文所題的方法在一輛自動車的導引上,證明本系統之可行。 An approach to the guidance of autonomous land vehicles (ALV's) by speciallandmark recognition using computer vision techniques for navigation incorridor environments is proposed. Several types of landmarks, such as ceilinglamps, plants, doors, trash cans, etc., are utilized for ALV guidance. Theceiling lamps in images are extracted and their edges are extracted toconstruct a straight line path for the ALV to follow by the line followingtechnique. The feature model of the special landmarks are learnedin advance and matched with the features extracted from navigation environmentimages. The desired path to a goal is specified by an ordered combination oflandmarks, and the required actions such as stops, turnings, etc., aredescribed by the meanings of these landmarks. A major type of feature usedin this study is extracted from the elliptical shape which fits best thecolor shape of the landmark. Furthermore, by the information of landmarksand lamps, a dead-reckoning method is used in corner turning.Lots of successful navigation experiments using a real ALV confirm the effectiveness of the proposed approach.