A Survey on Visual SLAM Algorithms Compatible for 3D Space Reconstruction and Navigation

Applications of the robots are increasing in routine for shop floor activity, transportation, and many other areas. While navigation, space reconstruction, and collision avoidance are the primary task of robots, not all simultaneous localization and mapping (SLAM) methods can be useful given that requirement of output is preferably in the specific form of 3D occupancy grid or point cloud in order to implement it on a real robot. This paper focuses on extensive study of conventional and deep learning based SLAM models that should be useful to create 3D output. In addition to that, we explored various available open source dataset considering provided ground truth convenient for evaluating 3D mapping and explain relevant evaluation criteria available in literature. Overall, this paper can be the first of its kind that focuses on all components of 3D SLAM systems.