Vision-based Global Localization Using Scale Invariant Keypoints

• Main Authors: Kai-jie Jhang, 張凱傑
• Other Authors: none
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/46850868783061175906

碩士 === 國立臺南大學 === 資訊工程學系碩士班 === 97 === 　　Due to the improvement of technology, handheld devices, such as personal digital assistants (PDA) and mobile phones, which are equipped with a camera and a global position system (GPS), are popularized. Therefore, the use of a personal navigation system for a person in an unfamiliar place is not only a dream. However, the accuracy of GPS depends much on the satellite signal which suffers from terrain occlusion or bad weather. Therefore, if we can make use of the camera and wireless communication functions of a handheld device, the use of a vision-based positioning system not only would be a low cost approach but also provides acceptable positioning accuracy. In this thesis, a vision-based positioning system consisting of the processes of perspective projection and 3D reconstruction between two viewpoints is proposed. By using of the scale invariant feature transformation algorithm to detect feature points on the test image and then creating feature descriptors, the matching process is able to search for the most similar image out from a pre-constructed database. The 3D coordinate and the relative transformation between the two matching images are computed based on the corresponding feature points. Finally, the location from which the image is taken can be derived from the relative transformation. The experimental results show that the proposed vision-based positioning system works well in various testing environments and applications.