碩士 / 國立臺灣科技大學 / 電機工程系 / 91 / In this thesis, we propose a region-of-interest-based scene recognition approach for the visual navigation of an indoor mobile robot. At the beginning, we capture the gray-level image from a single camera. Our algorithm does not modify the content of a raw image that directly serves as the input image. When the first raw image has been obtained, we will set the regions of interest and initialize the angles of both sides of a corridor. After that, we accomplish the edge detection in a corridor environment using two region modules. One is called a positive scoring region module. It is mainly used to calculate the degree of the position of a candidate pixel near the side of a corridor. The other is called a negative scoring region module that calculates the degree of how far a candidate pixel away from the side of a corridor. The fitness value of the candidate pixel can be obtained from the score of the former module subtracting that of the latter module. Finally, we sort the fitness values and select the better results from the whole candidate pixels. The side map of a corridor is composed of these selected candidate pixels. The module of feature points will be constructed from these selected candidate pixels. From this module, we can obtain the information for correcting the angles of both sides of a corridor. In addition, we propose the strategies for capturing scenes and avoiding collisions. The experimental results reveal that our approach is effective and efficient to recognize both sides of a corridor during the visual navigation of an indoor mobile robot.
Identifer | oai:union.ndltd.org:TW/091NTUST442079 |
Date | January 2003 |
Creators | Chien-Liang Chan, 詹健良 |
Contributors | Chin-Shyurng Fahn, 范欽雄 |
Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
Language | zh-TW |
Detected Language | English |
Type | 學位論文 ; thesis |
Format | 38 |
Page generated in 0.026 seconds