부산대학교 도서관

This paper mainly discusses the realization of affine Scale-Invariant Feature Transform (SIFT) real-time image processing method for “May” who is a home service robot designed and implemented by aiRobots laboratory. Firstly, a six-step method is proposed to create image database. Then the SIFT algorithm is adopted to detect features with this database, the error rate of features is reduced. Integration of the six-step method and SIFT can significantly decrease time for setting up the database. Thirdly, the object-recognition vision system is built up by parallel Affine-SIFT (ASIFT) algorithm, where the OpenGL is used to reduce the calculation time of affine projection and SiftGPU (Graphics Processing Unit) is adopted to accomplish the real-time operation speed of SIFT. Finally, the real-time grasping experiments demonstrate the feasibility and validity of the proposed scheme.