A Research of ORB Feature Matching Algorithm Based on Fusion Descriptor

Abstract

In the past few decades, visual SLAM has been successfully applied to technologies such as virtual reality and robot positioning. Among them, feature detection and matching technology is the key technology in SLAM. Aiming at the problems of large scale matching error and high mismatch rate of the binary description algorithm (Oriented fast and Rotated Brief (ORB)), an improved ORB feature matching algorithm in terms of scale and descriptors is proposed. Based on the binary description algorithm ORB, the algorithm constructs a pyramid-like scale space, and detects oFAST key points on each layer to improve the scale invariance of the algorithm. In terms of descriptors, the 128-bit improved FREAK description operator is used instead of the last 128 bits of the small variance in the rBRIEF description operator, which makes full use of image information to improve the matching accuracy and robustness. The experimental results show that the algorithm in this paper has greatly improved the feature matching rate and robustness in terms of scale change, rotation, and brightness change compared with the traditional ORB, and meets the requirements for fast and accurate matching of complex images.