A Comparison of Visual SLAM Algorithms ORB-SLAM3 and DynaSLAM on KITTI and TUM Monocular Datasets

  • Kesse Jonatas de Jesus UFSM
  • Mateus Osvaldo Klan Pereira UFSM
  • Leonardo Ramos Emmendorfer UFSM
  • Daniel Fernando Tello Gamarra UFSM

Resumo


This paper focuses on comparing two visual Simultaneous Localization and Mapping (vSLAM) algorithms: ORB-SLAM3 and DynaSLAM, utilizing simulations with a monocular camera. ORB-SLAM3 is an open-source library known for its monocular vSLAM capabilities and is an evolution of the well-regarded ORB-SLAM2. In contrast, DynaSLAM extends ORB-SLAM2 by incorporating Mask R-CNN for dynamic object detection, filtering, and segmentation. Both algorithms were tested and evaluated using sequences of monocular images from two popular datasets, KITTI and TUM RGB-D. The experiments demonstrate the efficiency of the vSLAM algorithms. The results reveal that DynaSLAM consistently outperforms ORB-SLAM3 in the majority of cases. Overall, this research contributes to the understanding of these vSLAM methods, providing insights into their performance and highlighting the advantages of DynaSLAM over ORB-SLAM3 in various scenarios.
Palavras-chave: vslam, DynaSLAM, ORB-SLAM3, monocular images
Publicado
06/11/2023
JESUS, Kesse Jonatas de; PEREIRA, Mateus Osvaldo Klan; EMMENDORFER, Leonardo Ramos; GAMARRA, Daniel Fernando Tello. A Comparison of Visual SLAM Algorithms ORB-SLAM3 and DynaSLAM on KITTI and TUM Monocular Datasets. In: CONFERENCE ON GRAPHICS, PATTERNS AND IMAGES (SIBGRAPI), 36. , 2023, Rio Grande/RS. Anais [...]. Porto Alegre: Sociedade Brasileira de Computação, 2023 . p. 109-114.