Options
Visual localization, mapping and reconstruction using edges
Citation Link: https://doi.org/10.15480/882.1282
Publikationstyp
Doctoral Thesis
Date Issued
2016
Sprache
English
Author(s)
Advisor
Title Granting Institution
Technische Universität Hamburg
Place of Title Granting Institution
Hamburg
Examination Date
2015-07-10
Institut
TORE-DOI
This dissertation proposes edge-based methods for stereo matching, visual localization, mapping and dense reconstruction. Particularly untextured scenes where feature point-based methods fail are investigated in order to develop complementary techniques that can still operate in minimal scene configurations with just two nonparallel line segments (i.e. polygonized edges). Using a new line segment-based visual odometry approach and bundle adjustment formulation, line segment maps are reconstructed in challenging scenes ranging from moderate complexity to virtually no texture. The proposed dense surface reconstruction from line segment maps completes the set of algorithms allowing for scenarios such as real-time robotic exploration.
Subjects
Localization
Mapping
Line SLAM
Visual Odometry
Line Segments
Dense Surface Reconstruction from Line Map
Sparse Stereo Edge Matching
Line Bundle Adjustment
DDC Class
620: Ingenieurwissenschaften
Loading...
Name
Dissertation_Jonas_Witt_compressed.pdf
Size
4.31 MB
Format
Adobe PDF
Loading...
Name
Dissertation_Jonas_Witt.pdf
Size
35.77 MB
Format
Adobe PDF