The current SLAM algorithms cannot work without assuming rigidity. We propose the first real-time tracking thread for monocular VSLAM systems that manages deformable scenes. It is based on top of the Shape-from-Template (SfT) methods to code the scene deformation model. Our proposal is a sequential method that manages efficiently large templates, i.e. deformable maps estimating at the same time the camera pose and deformation. It also can be relocated in case of tracking loss. We have created a new dataset to evaluate our system. Our results show the robustness of the method in deformable environments while running in real time with errors under 3% in depth estimation.
CITATION STYLE
Lamarca, J., & Montiel, J. M. M. (2019). Camera tracking for SLAM in deformable maps. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11129 LNCS, pp. 730–737). Springer Verlag. https://doi.org/10.1007/978-3-030-11009-3_45
Mendeley helps you to discover research relevant for your work.