OPDS-SLAM: A Semantic Visual SLAM Using Object-Plane Feature Towards Dynamic Environment

A novel visual SLAM method named OPDS-SLAM is proposed to address the dynamic indoor environments and improve the robot’s understanding of the environment. First of all, OPDS-SLAM utilizes YOLOv11 to identify human-beings in the scene and combines the depth information to remove human-beings’ features. Second, the spherical coordinates are used to represent the planar information in the scene, and the stability of the system is improved by constructing a backend optimization with planar features. Third, the objects are represented by the ellipsoids, and the planar constraints are added to enhance the speed and accuracy of object construction. Finally, the experiments conducted on the TUM RGB-D dataset and a self-built dataset demonstrate that OPDS-SLAM can remove moving human-beings’ feature in the indoor environments in real-time. Meanwhile a semantic map containing object-plane information is constructed.