Open AccessVisual odometer method based on improved ORB feature
Author(s) -
Xiaonan Wang,
Feifei Liu,
Yaxin Xue
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1920/1/012110
Subject(s) - ransac , artificial intelligence , computer science , quadtree , computer vision , feature (linguistics) , pattern recognition (psychology) , orb (optics) , robustness (evolution) , feature extraction , matching (statistics) , mathematics , image (mathematics) , philosophy , linguistics , biochemistry , chemistry , statistics , gene
The original ORB feature detection algorithm has many problems, such as uneven distribution of feature points, many feature mismatches and poor robustness. A visual odometer method based on improved ORB is proposed. First of all, in order to solve the problem of low extraction efficiency and uneven distribution in the process of key points extraction, an improved quadtree algorithm is proposed which set an appropriate quadtree depth for the image pyramid layer to improve calculation efficiency and filter key points. Secondly, aiming at the problem of mismatching in the feature matching process of adjacent frames, the PROSAC algorithm improved based on RANSAC is used to sort all the sampling points according to the matching quality, and the matching point pairs with good quality are selected to improve the matching accuracy. The experimental results shows that the improved visual odometer method has higher accuracy and less calculation time than the original algorithm.