DOC-SLAM: Robust Stereo SLAM with Dynamic Object Culling

Lin Lyu, Yan Ding*, Yating Yuan, Yutong Zhang, Jinpeng Liu, Jiaxin Li

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

5 Citations (Scopus)

Abstract

To improve the accuracy of estimating camera trajectory in dynamic scenes, this paper proposes Dynamic Object Culling SLAM(DOC-SLAM), a stereo SLAM system that achieves good performance by culling actual moving objects in highly dynamic environments. DOC-SLAM combines the semantic information from panoptic segmentation with the point features from optical flow together to detect potential moving objects. And a moving consistency check module is designed to determine and remove the feature points in objects which are in motion so as to accomplish dynamic objects culling. Besides, for enhancing the robustness of our system, we devise a key point supplement strategy to provide sufficient and reliable key points for tracking. Meanwhile, the trajectory and landmarks are generated for localization and mapping of robots. The experimental evaluation on public datasets demonstrates that our DOC-SLAM can fit highly dynamic scenes.

Original languageEnglish
Title of host publication2021 International Conference on Automation, Robotics and Applications, ICARA 2021
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages258-262
Number of pages5
ISBN (Electronic)9780738142906
DOIs
Publication statusPublished - 4 Feb 2021
Event2021 International Conference on Automation, Robotics and Applications, ICARA 2021 - Virtual, Prague, Czech Republic
Duration: 4 Feb 20216 Feb 2021

Publication series

Name2021 International Conference on Automation, Robotics and Applications, ICARA 2021

Conference

Conference2021 International Conference on Automation, Robotics and Applications, ICARA 2021
Country/TerritoryCzech Republic
CityVirtual, Prague
Period4/02/216/02/21

Keywords

  • SLAM system
  • dynamic object culling
  • panoptic segmentation
  • trajectory estimation

Fingerprint

Dive into the research topics of 'DOC-SLAM: Robust Stereo SLAM with Dynamic Object Culling'. Together they form a unique fingerprint.

Cite this