Skip to main navigation Skip to search Skip to main content

GCP-SLAM: LSD-SLAM with Learning-Based Confidence Estimation

  • School of Computer Science and Technology, Harbin Institute of Technology
  • School of Mechatronics Engineering, Harbin Institute of Technology

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

Abstract

Astonishing progress has been made in direct monocular SLAMs in the last few years. However, most direct methods, such as large-scale direct monocular SLAM (LSD-SLAM), usually have lower camera localization accuracy than feature-based methods. To tackle this issue, this paper suggests a novel LSD-SLAM model, i.e., GCP-SLAM, by incorporating with learning-based confidence estimation. A regression forest method is used to estimate confidence and select ground control points (GCPs). The estimated confidence and GCPs are then exploited for improving depth estimation and camera localization, respectively. Experiments show that GCP-SLAM is more reliable in tracking and relocalization than LSD-SLAM.

Original languageEnglish
Title of host publicationImage and Video Technology - 8th Pacific-Rim Symposium, PSIVT 2017, Revised Selected Papers
EditorsCarlos Hitoshi, Manoranjan Paul, Qingming Huang
PublisherSpringer Verlag
Pages262-275
Number of pages14
ISBN (Print)9783319757858
DOIs
StatePublished - 2018
Externally publishedYes
Event8th Pacific-Rim Symposium on Image and Video Technology, PSIVT 2017 - Wuhan, China
Duration: 20 Nov 201724 Nov 2017

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10749 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference8th Pacific-Rim Symposium on Image and Video Technology, PSIVT 2017
Country/TerritoryChina
CityWuhan
Period20/11/1724/11/17

Keywords

  • Ground control points (GCPs)
  • Localization
  • Monocular SLAM
  • Random forest

Fingerprint

Dive into the research topics of 'GCP-SLAM: LSD-SLAM with Learning-Based Confidence Estimation'. Together they form a unique fingerprint.

Cite this