MISD-SLAM: Multimodal Semantic SLAM for Dynamic Environments

Figure 1

The overview of MISD-SLAM. The raw RGB image is input to perform ORB feature extraction and instance segmentation for semantic information simultaneously. Then, the system removes dynamic features and continues tracking thread to calculate camera pose. Finally, semantic 3D map is built using the data of depth image, camera pose, and semantic information in an independent thread.