G2P-SLAM: Generalized Grouping and Pruning Method for RGB-D SLAM in Low-Dynamic Environments

Abstract

In this paper, we propose a generalized grouping and pruning method for RGB-D SLAM in low-dynamic environments. The current grouping and pruning algorithm works well in low-dynamic environments. However, it fails when high-dynamic objects are dominant, as it also removes the nodes with low-dynamic landmarks which are true constraints within grouped nodes. As a solution we suggest to group nodes with information from feature-tracking, so that greater number of similar features can be shared among the grouped nodes. Likelihood based on the Mahalanobis distance is used to detect dynamic features, and constraints between node groups are grouped to filter out the constraints from remaining lowdynamic objects. The system’s performance is verified and compared with other algorithms.