# Dynamic Dense RGB-D SLAM using Learning-based Visual Odometry

arXiv: https://arxiv.org/abs/2205.05916 · HTML: https://mlsanity.com/paper/2205.05916/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 7 of its 20 references that are in the map.
- First posted on arXiv 12 May 2022.

## Builds on (most cited first, 7 of 7)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume](https://mlsanity.com/paper/1709.02371/) (2017)
- [DynaSLAM: Tracking, Mapping and Inpainting in Dynamic Scenes](https://mlsanity.com/paper/1806.05620/) (2018)
- [TartanVO: A Generalizable Learning-based VO](https://mlsanity.com/paper/2011.00359/) (2020)
- [MID-Fusion: Octree-based Object-Level Multi-Instance Dynamic SLAM](https://mlsanity.com/paper/1812.07976/) (2018)
- [Co-Fusion: Real-time Segmentation, Tracking and Fusion of Multiple Objects](https://mlsanity.com/paper/1706.06629/) (2017)
- [Learning Rigidity in Dynamic Scenes with a Moving Camera for 3D Motion Field Estimation](https://mlsanity.com/paper/1804.04259/) (2018)

## Abstract (arXiv)

We propose a dense dynamic RGB-D SLAM pipeline based on a learning-based visual odometry, TartanVO. TartanVO, like other direct methods rather than feature-based, estimates camera pose through dense optical flow, which only applies to static scenes and disregards dynamic objects. Due to the color constancy assumption, optical flow is not able to differentiate between dynamic and static pixels. Therefore, to reconstruct a static map through such direct methods, our pipeline resolves dynamic/static segmentation by leveraging the optical flow output, and only fuse static points into the map. Moreover, we rerender the input frames such that the dynamic pixels are removed and iteratively pass them back into the visual odometry to refine the pose estimate.