A fast visual feature matching algorithm in multi-robot visual SLAM

Abstract

To reduce the feature matching time in visual based multi-robot Simultaneous Localization and Mapping (SLAM), a feature matching algorithm based on map environment is proposed in this paper. The key idea of our algorithm is to establish feature libraries by classifying the collected features into two categories during the mobile process of every sub-robot. Then all features are matched based on the categories so that the invalid feature matching time will be reduced. At last, experiment is conducted to verify the performance of proposed algorithm. In comparison with traditional BoW method, its feature matching time is reduced by 20% at no expense of accuracy.