Hits:

Indexed by:会议论文

Date of Publication:2017-01-01

Included Journals:Scopus、EI、CPCI-S

Volume:10262

Page Number:336-344

Key Words:Image matching; RTM; Binary descriptors; FREAK; rBRIEF

Abstract:Geometric constraints have been widely applied to image matching to gain additional advantages over feature points. A rapid triangle matching (RTM) algorithm was such an algorithm for matching triangles formed by three feature points using 38-dimensional floating-point descriptors. The RTM was faster than SIFT, but it was still hard to meet the real-time requirement. As such, we designed two kinds of binary descriptors including FREAK and rBRIEF used in ORB to replace the floating-point descriptors of RTM, and compared the improved RTM algorithms with original RTM algorithm and SIFT based on simulated and actual binocular images. The results demonstrate that our algorithms greatly improve the speed as well as the precision and matching score. Furthermore, our algorithms can match additional points compared to SIFT and the original RTM when applied to structural scenes.