Spatial structure comparison based RGB-D SLAM in dynamic environments

Simultaneous Localization and Mapping with RGB-D (RGB-D SLAM) is a hot research issue in the robotic field. How to remove the impact of dynamic objects on RGB-D SLAM is still a challenging task. To address this issue, an improved RGB-D SLAM based on spatial structure comparison is proposed in this paper. In the positioning stage of the proposed RGB-D SLAM, feature point match pairs are divided into two groups based on depth information, namely depth-valid and depth-misaligned. By comparing the spatial structure matrices of the two three-dimensional point sets, static feature point matched pairs are identified. Furthermore, the obtained pose is used to remove depth-misaligned dynamic feature point matched pairs, aiming to enhance the performance of back-end optimization. During the map processing stage, a novel map point method incorporating distance detection and handling coordinate errors is employed. And a duplicate point removal strategy is presented to eliminate redundant map points in the reference frame. Finally, the proposed RGB-D SLAM is tested on the public dynamic TUM RGB-D dataset. The experimental results demonstrate that the proposed RGB-D SLAM has superior performance, compared to the state-of-the-art RGB-D SLAM methods under dynamic environments.