DPF-SLAM: Dense semantic SLAM based on dynamic probability fusion in dynamic environments

Xin Liu, Shuhuan Wen (Lead / Corresponding author), Mingxing Yuan, Pengjiang Li, Yongjie Zhao, Luigi Manfredi

Research output: Chapter in Book/Report/Conference proceedingConference contribution

5 Downloads (Pure)

Abstract

With the development of simultaneous localization and mapping (SLAM) technology, Dynamic SLAM has been a challenging research topic. This paper presents a dynamic probability fusion SLAM (DPF-SLAM) algorithm, which adds a semantic segmentation thread and a dense reconstruction thread to ORB-SLAM2. We integrate dynamic prior probability obtained by semantic segmentation with dynamic probability obtained by dynamic point detection, which can decrease the effect of dynamic objects on the localization accuracy and meanwhile achieve the dense reconstruction of a static background. We evaluate DPF-SLAM system on the public TUM RGBD dataset. The experimental results show that the proposed algorithm has better localization accuracy than DS-SLAM and ORB-SLAM2, which are two relatively new algorithms in this area, and can obtain a good dense reconstruction effect. Moreover, through the performance comparison with our previous work, it is found that the algorithm speed and positioning accuracy are improved.

Original languageEnglish
Title of host publication2022 IEEE International Conference on Real-Time Computing and Robotics, RCAR 2022
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages360-365
Number of pages6
ISBN (Electronic)9781665469838
DOIs
Publication statusPublished - 5 Sep 2022
Event2022 IEEE International Conference on Real-Time Computing and Robotics, RCAR 2022 - Guiyang, China
Duration: 17 Jul 202222 Jul 2022

Publication series

Name2022 IEEE International Conference on Real-Time Computing and Robotics, RCAR 2022

Conference

Conference2022 IEEE International Conference on Real-Time Computing and Robotics, RCAR 2022
Country/TerritoryChina
CityGuiyang
Period17/07/2222/07/22

Fingerprint

Dive into the research topics of 'DPF-SLAM: Dense semantic SLAM based on dynamic probability fusion in dynamic environments'. Together they form a unique fingerprint.

Cite this