本論文主要目的為使用圖樣辨識的方法去控制自走車的移動方向進而達到導航的目的。本論文在一部自走車上裝設一台攝影機，利用這部攝影機去搜尋在地面上的圖樣。自走車會自行去靠近圖樣、對正圖樣、辨識圖樣然後做出圖樣所代表的動作。當執行完此圖樣的動作，自走車會接著搜尋下一個圖樣，然後靠近圖樣、對正圖樣、辨識圖樣，如此循環下去就可以達到自走車的導航控制。 在圖樣偵測部分是利用顏色偵測技術去搜尋需要辨識的圖樣區塊，並且把圖樣從原本複雜的背景影像中抽離出。在圖樣擷取部分則是利用輪廓追蹤技術來取出可以用來判斷特徵的符號，此技術也可以順便濾除掉在背景抽離時所殘留的雜訊。在取得可以用來判斷特徵的圖樣之後，使用多變量分析中的二維線性鑑別分析法來取得圖樣的特徵參數，利用這個特徵參數去做識別的動作，將辨識出來的結果控制自走車的動作。 在估測圖樣與自走車之間的相對距離與位置，本論文設計了一個倒傳遞神經網路用來執行反透視法的功能。利用倒傳遞神經網路的函數近似功能，可以在無法得知攝影機內部參數的情況下，只要給予足夠的輸入與輸出，就可以逼近攝影機內部非線性的影像投影函數。自走車會根據自己與圖樣之間的相對距離與位置來去評估該如何移動自己進行靠近圖樣、對正圖樣的動作。 本論文使用了影像辨識技術與倒傳遞神經網路，來達到智慧型控制自走車自動導航的目的。The main purpose of this thesis is to use pattern recognition methods to control the autonomous mobile robot and then achieve the purpose of navigation. The thesis installed a video camera in the autonomous mobile robot and used it to search for patterns on the ground. The autonomous mobile robot will go ahead to approach the pattern, identify it and then perform the action for which the pattern represented. When this pattern action completed, the next pattern is sought and the action will continue, so the circle continues to achieve the navigation control of the autonomous mobile robot. As for the pattern detection, color detection techniques are used to search for patterns and identify the block, and it will separate the image of pattern from the original complex background. As for the capturing pattern section, contour tracing techniques are used to extract the pattern that can be used to determine the features. This technique also filtered out residual noise from the background in the detached way. After the access pattern is used to determine characteristics, two-dimensional linear discrimination analysis of method is employed to obtain the characteristic parameters of pattern. By using the characteristic parameters to recognize the pattern, the results will be identified to control an autonomous mobile robot. The estimation of the relative distance and the location between autonomous mobile robot and the pattern section, this thesis designs a back-propagation neural network to execute on inverse perspective mapping. Using the back-propagation neural network function approximation can give enough input and output to approximate the camera within the nonlinear images projection function in the situation of not knowing the camera internal parameters of the case. According to the relative distance between the autonomous mobile robot and the pattern, the autonomous mobile robot will evaluate how to move, approach the pattern and correct the action automatically. The thesis uses the image recognition technology and the back-propagation neural network, to achieve the navigation purposes of intelligently controlling autonomous mobile robot.
