Single Platform of FPGA-Based SIFT for Visual Recognition and Landmark Searching of a Mobile Robot

• Main Authors: Zhong-yan Yang, 楊忠諺
• Other Authors: Chih-lyang Hwang
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/85334345353059798320

碩士 === 國立臺灣科技大學 === 電機工程系 === 102 === It is known that feature extraction is a fundamental part to recognize objects in an image processed. For the purpose of robustly extracting these features, the demanded computation using software approach is difficult to obtain on-line applications. In this thesis, an FPGA-based scale invariant feature transform (SIFT) is implemented to accelerate the recognition and description of these features. The extinguisher and garbage can are chosen as specific landmarks and they are also pre-trained to accomplish their corresponding feature vectors stored in the FPGA platform. The matching criterion using linear search is then constructed to evaluate the successful recognition rate of the specific landmarks. The strategy for the search of the chosen landmarks through a mobile robot is designed and evaluated by different conditions. Finally, the corresponding experiments are given to validate the effectiveness and robustness of the proposed methodology.