关键词: 特征检测, 二进制描述子, 尺度不变性, 旋转不变性, 实时性

Abstract: Concern the problem that LATCH descriptor lack of scale invariance and its rotation invariance depends upon feature detector, a integrated algorithm was proposed based on ORB-LATCH. The proposed algorithm adopt FAST to detecte corner feature on the scale space of image pyramid and intensity centroid method of ORB to obtain orientation compensation, and on this basis use LATCH to describe the feature. Experiments indicate that the proposed algorithm can detect and describe the feature quickly and obtain scale and rotation invariance, and its recall is better than ORB and HARRIS-LATCH algorithm, its matching inner rate is higher than the original binary descriptors 4%. In conclusion, the proposed algorithm can close the performance gap between and histogram based descriptors such as SIFT and SURF with the real-time property to deal image sequence exactly.

Key words: feature detector, binary descriptors, scale invariance , rotation invariance, real-time property