MISD-SLAM: Multimodal Semantic SLAM for Dynamic Environments

Figure 4

The flowchart of dynamic pixel removal. The raw RGB image is firstly utilized to instance segmentation to remove the pixels located on dynamic objects. Then, the remaining pixels would be combined with the depth image to generate clusters using -means method. Under multiview geometry constraint, points can be divided into dynamic and static. And the motion state of each cluster is determined by voting method.