Advanced

Efficient and robust map building from a downward looking camera with loop closing

Brange, Elias LU (2016) In Master's Theses in Mathematical Sciences FMA820 20161
Mathematics (Faculty of Engineering)
Abstract
Since good cameras are becoming cheaper and embedded systems are getting more powerful, using cameras as the only sensor is getting more popular for visual navigation of robots. By using a robot to explore a flat surface a map can be gradually generated using the images taken by a downward looking camera. The thesis focus lies primarily on offline map building, where complete datasets are first gathered and then analysed without real-time requirements.

An algorithm to build the map is developed and tested on various simulated and real datasets. Images are aligned sequentially to estimate relative robot translation and rotation between image frames. Such estimations are prone to drifting errors, errors which will be minimized by finding... (More)
Since good cameras are becoming cheaper and embedded systems are getting more powerful, using cameras as the only sensor is getting more popular for visual navigation of robots. By using a robot to explore a flat surface a map can be gradually generated using the images taken by a downward looking camera. The thesis focus lies primarily on offline map building, where complete datasets are first gathered and then analysed without real-time requirements.

An algorithm to build the map is developed and tested on various simulated and real datasets. Images are aligned sequentially to estimate relative robot translation and rotation between image frames. Such estimations are prone to drifting errors, errors which will be minimized by finding loop closings in the image sequence. A loop closing is found whenever the algorithm detects that the robot has entered a previously mapped area.

The algorithm shows promising results on both real and simulated datasets. However, the time complexity for finding loop closings is far from meeting a real-time requirement. Further work includes optimization of the algorithm to exploit the structure of the system, as well as implementing an efficient search method to find loop closings in the image sequence. (Less)
Please use this url to cite or link to this publication:
author
Brange, Elias LU
supervisor
organization
course
FMA820 20161
year
type
H2 - Master's Degree (Two Years)
subject
publication/series
Master's Theses in Mathematical Sciences
report number
LUTFMA-3305-2016
ISSN
1404-6342
other publication id
2016:E45
language
English
id
8893989
date added to LUP
2016-11-16 10:43:54
date last changed
2016-11-16 10:43:54
@misc{8893989,
  abstract     = {Since good cameras are becoming cheaper and embedded systems are getting more powerful, using cameras as the only sensor is getting more popular for visual navigation of robots. By using a robot to explore a flat surface a map can be gradually generated using the images taken by a downward looking camera. The thesis focus lies primarily on offline map building, where complete datasets are first gathered and then analysed without real-time requirements. 

An algorithm to build the map is developed and tested on various simulated and real datasets. Images are aligned sequentially to estimate relative robot translation and rotation between image frames. Such estimations are prone to drifting errors, errors which will be minimized by finding loop closings in the image sequence. A loop closing is found whenever the algorithm detects that the robot has entered a previously mapped area. 

The algorithm shows promising results on both real and simulated datasets. However, the time complexity for finding loop closings is far from meeting a real-time requirement. Further work includes optimization of the algorithm to exploit the structure of the system, as well as implementing an efficient search method to find loop closings in the image sequence.},
  author       = {Brange, Elias},
  issn         = {1404-6342},
  language     = {eng},
  note         = {Student Paper},
  series       = {Master's Theses in Mathematical Sciences},
  title        = {Efficient and robust map building from a downward looking camera with loop closing},
  year         = {2016},
}