CAJ | 학술논문

在视觉引导的机器人抓取技术中，需要快速提取不同视角图像的特征点集，实现基于特征的姿态估计。传统的特征点提取算法十分耗时，并难以在基于DSP的嵌入式图像处理平台上实现。针对工业应用场景，提出了一种基于简化SURF和不变矩的特征点提取算法，引入了一个4维的局部不变矩作为特征描述子。算法针对DSP 平台进行了优化，测试达到了7帧／s的运行速度，并满足一定程度的光照和仿射不变性。
재시각인도적궤기인조취기술중，수요쾌속제취불동시각도상적특정점집，실현기우특정적자태고계。전통적특정점제취산법십분모시，병난이재기우DSP적감입식도상처리평태상실현。침대공업응용장경，제출료일충기우간화SURF화불변구적특정점제취산법，인입료일개4유적국부불변구작위특정묘술자。산법침대DSP 평태진행료우화，측시체도료7정／s적운행속도，병만족일정정도적광조화방사불변성。
Quick extraction of the image feature point set in different perspectives was needed in robust grasp technology to re -alize pose estimation based on characteristics .Traditional image interest point extraction methods are time-consuming,and it is hard to perform in real-time and duplicate to implement on DSP embedded platform .Aiming at industrial applications , a new interest point extraction method based on SURF and moment invariants was proposed .The local moment invariants were introduced to resist illumination change and affine transform ,which constructs only a 4-dimensional descriptor .And the DSP-specified optimization was conducted systematically .Experiments show that the algorithm can process nearly 7 images per second on the DSP platform ,with the stability under changing illumination and affine transformation .