臺灣大學: 電機工程學研究所連豊力楊峻瑜Yang, Jun-YuJun-YuYang2013-03-272018-07-062013-03-272018-07-062012http://ntur.lib.ntu.edu.tw//handle/246246/253881對於一個行動機器人，在一個未知環境中能夠正確並自主地移動是必備的能力。定位、消失點偵測和地面區域偵測的問題都和這能力密切相關。如果機器人能夠知道自身位置，則它可以移動的更準確。此外，有了對消失點及地面區域的認知，機器人可以自主且安全地探索環境。 視覺感測器，例如:攝影機，是個非常有吸引力的裝置，由於擁有非常豐富的資訊。比起單眼相機，全方位攝影機提供了更大的視野以及完整的周圍環境資訊。本論文提出了對於定位與偵測消失點及地面的方法，而這些方法都是基於全方位視覺。 對於定位問題來說，定位問題被分成兩個部份來討論:位移估測及轉動估測。機器人的轉動角度可以利用兩張全景影像做垂直線比對而得到。機器人位移則是利用了一個基礎的裝置，馬達編碼器，經過計算而求得。 對於消失點偵測問題來說，利用Sobel邊界偵測方法，使得全景影像中的邊界方向資訊可以被獲得。接下來，邊界方向資訊被以兩種不同方法做分析，進而找出消失點。因此，本論文針對消失點偵測提出了兩種方法。 對於地面偵測問題來說，我們提出了一個模仿雷射測距儀尋找最近障礙物的系統，並使用了Canny邊界偵測。為了了解地面區域所在方向，消失點位置被做為是該地面偵測系統的已知條件。最後，為了在導航過程中偵測路面的便利性，本論文提出一個結合消失點偵測及路面偵測的整合方法。 在許多情況下的垂直線配對模擬結果驗證了該演算法計算旋轉角度的可行性。此外，針對三種任務(即定位、消失點偵測、地面偵測)的方法在各種室內環境運行，如圖書館、走廊。實驗結果將被呈現並和雷射測距儀的結果相比較。For a mobile robot, it is necessary to have the ability to move accurately and autonomously in an unknown environment. The localization, vanishing point detection, and floor region detection problem all have a close relation to this ability. If the robot is able to localize itself, it can move accurately. Besides, the robot can explore the environment autonomously and securely with knowing the vanishing points and the floor region. Vision sensors such as camera are attractive devices since they are information-rich. Compared with monocular cameras, omnidirectional cameras provide a larger field of view and complete information about the surroundings. In this thesis, the methods to solve the tasks related to localization, vanishing point detection, and floor region detection are proposed. All the methods are based on the omnidirectional vision. For localization problem, this problem is divided into rotation measurement and translation measurement. The rotation angle of robot is measured by vertical line matching (VLM) using two consecutive omnidirectional images. The translation distance is calculated by a basic device, namely encoder. For vanishing point detection problem, the orientation information in the panoramic image is computed by the Sobel edge detection. Next, the orientation information is analyzed in two ways: local dominant orientation signature (LDOS) and slope trend. Therefore, two different approaches for detecting vanishing points are presented. For floor region detection problem, we propose a method imitating the way range finders detect the closest objects with the Canny edge detection. To realize the directions of floor, the vanishing point information serves as the prior knowledge. Finally, for the purpose of detecting floor more conveniently when navigation, an integrated approach combined with vanishing point detection and floor region detection is provided. The results of simulation of vertical line matching in different cases validate the feasibility of the algorithm to compute the rotation angle. In addition, the methods to solve the three tasks (i.e., localization, vanishing point detection, and floor region detection) are performed in various indoor environments such as library and corridor. The experiment results will be shown and compared with the results from laser range finder.9392106 bytesapplication/pdfen-US全方位攝影機全景影像視覺定位系統消失點偵測地面區域偵測omnidirectional camerapanoramavision localization systemvanishing point detectionfloor region detectionthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/253881/1/ntu-101-R99921061-1.pdf