Threedimensional Object Mapping and Reconstruction through Stereoscopic Image Processing

1
Three-dimensional Object Mapping and
Reconstruction through Stereoscopic Image
Processing
Jama Mohamed and Julian WarchallTJHSST 2009
2
Goal

• To develop a system that
• Uses two cameras
• Senses depth of all objects in environment
• Creates a vertex map from depth data

3
Background

• Current efficient vertex-mapping systems are
• Too expensive
• Too bulky
• Too complicated

4
Checkpoints

• Create a working algorithm for depth perception
  using Matlab and static images.
• Integrate two cameras into the system for
  livevideo feeds.
• Parallelize the algorithm and distribute
  acrossmultiple Beagle Boards for faster
  processing
• Write algorithm to create vertex maps from
  depth data

5
Algorithm

• Looking for disparity
• Disparity corresponds toobject's distance
  greaterdisparity shorter distance
• Calculating pixel disparity over scan-lines

6
Algorithm

• Think of color as a function of x-position
• Two spheres

7
Algorithm

• Disparity finding

8
Algorithm

• Features
• Advantage Per-scan-line basis for processing
• Disadvantage Cameras have to be at exactlysame
  y-position

9
Parallelization

• Algorithm lends itself to distribution
• Beagle Board

10
The End

• Questions?