Omnidirectional Vision for Robot Navigation

1
Omni-directional Vision for Robot Navigation

• Niall Winters, Jose Gaspar, Gerard Lacey, Jose
  Santos-Victor
• Presented by
• Hamed Masnadi-Shirazi

2
General Paper Overview

• Omni-directional Camera/Birds Eye View
• Topological Navigation
• a) Position Estimation.
• b) Illumination Changes.
• c) Occlusion.
• d) Corridor Following.
• Visual Path Finding
• a) Feature Tracking.
• Experimental Results
• Topological Map Navigation
• Visual Path Following
• Integrated Experiment
• Future Work

3
Omni-directional/Birds Eye View

• The Workings Of The Omni-directional Camera

4
Model Parameter Estimation

• How to estimate all the camera parameters
  needed to find the projection operator.

5
Birds Eye View

• A method to produce the Birds Eye View of
  the ground plane from the Panoramic image.

6
Navigation Using Topological Maps

• Nodes Identified by Images
• Links Identified by Image Features
• Position Estimation using Eigenspace
  representation. (Next Slide)

7
EigenSpace Representation of Images
8
Canceling Illumination Changes

• Matching is achieved using the Hausdorff Fraction
  and Distance
• http//www.cs.ucla.edu/hljin/cs269/CS269-syllabus
  .html

9
Hausdorff Distance?

where a and b are points of sets A and B
respectively http//http//cgm.cs.mcgill.ca/godf
ried/teaching/cg-projects/98/normand/main.html
10
Occlusion and Corridor Following

• Using the Eigenspace approach the closest image
  is correctly determined despite occlusion
• Bounding boxes and line fitting are used to
  follow corridors.

11
Visual Path Finding

• Find and Track Corners
• Estimate Relative Position and Orientation
• Minimize Distance and Orientation Errors

12
Experimental Results

• Topological Navigation Results
• Visual Path Finding Results
• Integrated Experiment
• Compare to Odometer

13
Future Work

• Robot Navigation for Complex Environments
• Automatic Extraction of Landmarks