SLAM Algorithm Integrating Semantic and Point Cloud Information

The SLAM algorithm assumes that the collected scene is static during the process of localization and mapping. However, it is inevitable to encounter interference from dynamic objects such as pedestrians in the actual usage process, which will seriously affect the stability of the system. Although there is no mismatch, the matching pairs on dynamic objects do not conform to the geometric relationship between the two views, which can affect the accuracy of the pose solution. At the same time, the movement of dynamic objects will also affect the matching effect. In addition, introducing feature points on dynamic objects into the global map will affect optimization performance and interfere with the matching of visual localization. To address these issues, this paper proposes a SLAM algorithm that integrates semantic and point cloud information. Firstly, the input images are performed object detection, then dynamic points are deleted in the detection area by using multi-view geometric relation, and finally static points and semantic information are used to obtain the pose solution. Simulation and experimental results show that the proposed algorithm can effectively improve the robustness of the SLAM system in dynamic environments.