Rückert D, Innmann M, Stamminger M (2019)
Publication Language: English
Publication Type: Conference contribution, Original article
Publication year: 2019
DOI: 10.1109/ismar-adjunct.2019.00-15
We present FragmentFusion, a real-time dense reconstruction
pipeline that combines sparse camera tracking with image-space
volumetric fusion. The tracking is based on ORB-SLAM, which
constructs and optimizes a sparse global map of 3D points and
keyframes. We transform each of these keyframes into decimated
meshes and render them from the estimated viewpoint. Fusion is
performed in the pixel shader by exploiting atomic operations on a
packed data structure. This eliminates the need of a 3D voxel grid
making FragmentFusion very flexible at large scenes and varying
scales. Moreover, since all keyframes are fused on-the-fly, we can
use bundle adjustment and loop-closure without expensive volu-
metric re-integration. FragmentFusion is lightweight in terms of
compute power and memory consumption. It can easily fuse sev-
eral hundreds of keyframes in real time, in a quality comparable
to other approaches. We achieve real-time frame rates on a note-
book at around 20% CPU and GPU utilization and low memory
consumption.
APA:
Rückert, D., Innmann, M., & Stamminger, M. (2019). FragmentFusion: A Light-weight SLAM Pipeline for Dense Reconstruction. In Proceedings of the ISMAR 2019. Peking, CN.
MLA:
Rückert, Darius, Matthias Innmann, and Marc Stamminger. "FragmentFusion: A Light-weight SLAM Pipeline for Dense Reconstruction." Proceedings of the ISMAR 2019, Peking 2019.
BibTeX: Download