| Summary: | 碩士 === 國立臺灣科技大學 === 電機工程系 === 103 === In this thesis, an FPGA-based feature detection and matching (FDM) algorithm is first implemented to accelerate the recognition and description of the features of landmarks. For the localization of mobile robot (MR), some selected landmarks (e.g., extinguisher, garbage can with suitable Chinese characters) are first trained by the proposed FDM algorithm to accomplish their corresponding feature vectors stored in the FPGA platform. These landmarks are unknown distribution in an environment for the localization of MR. After the matching with the pre-trained feature vector of landmarks, the 2D coordinate of the corresponding landmark is estimated and then stored into a map, so that it can be as a reference map with known landmarks for the visual navigation of an MR. Since the location of landmarks is unknown, an on-line searching strategy of MR is designed. For fully development of the proposed platform, the PID control of two motors using kinematics of MR, the removal of unsuitable feature points to improve the accurate localization of landmark, and the searching strategy of specific landmarks are arranged in the NIOS with the software manner. On the other hand, image capturing, processing, extraction of feature points, feature descriptor, and landmark matching are executed by FPGA hardware. Finally, the corresponding experiments are given to validate the effectiveness and practicality of the proposed methodology.
|
|---|
