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.
@article{arxiv.2205.05916,
title = {Dynamic Dense RGB-D SLAM using Learning-based Visual Odometry},
author = {Shihao Shen and Yilin Cai and Jiayi Qiu and Guangzhao Li},
journal= {arXiv preprint arXiv:2205.05916},
year = {2023}
}