An incremental loop closure detection method based on depth information for indoor dynamic scenes

摘要

We present a RGB-D based place recognition algorithm methods for large-scale dynamic indoor environment. In contrast with other methods, our methods focus on the scene that dynamic objects are remarkable and avoid the unnecessary calculate expenses on the whole image sequences. In order to prohibit the influence of dynamic object, the depth information was applied to discriminate the image foreground information and filter it out. We conducted our methods on the public dataset and tested on the real environment. The results demonstrate that the proposed methods can effectively improve the loop-closure system in indoor dynamic environments.