DR-Fusion: Dynamic SLAM 3D Reconstruction Method of Production Workshop

During the movement process and autonomous navigation according to different production tasks, reconstruction the 3D model of the workshop scene is considered to be the basic capability of workshop mobile robots. However, the production workshop is a dynamic environment. These dynamic objects in the workshop not only reduce the positioning accuracy of the robot, but also will be contained in the reconstructed map. In addition, since the production workshop belongs to large scene, the built model consumes the memory of the mobile device extremely. In response to the above problems, this paper proposes a dynamic model segmentation criterion to eliminate dynamic objects in the process of position and scene model recovery. At the same time, the built model is sparsely stored to reduce resource consumption. In order to prove the effectiveness of this method of DR-Fusion, experiments are carried out on the TUM public datasets and the real workshop environment. The experimental results show that this method can not only improve the mapping effect by improving the positioning accuracy, but also can effectively eliminate the dynamic objects in the scene model and store the model sparsely; finally, a geometric model only containing static objects in the scene can be obtained.