Vision Subsystem of Impossibles AIBO 4legged Fotballers

1
Vision Sub-system of Impossibles AIBO 4-legged
Fotballers

• Kianoosh Mokhtarian

Computer Engineering Department Sharif University
of Technology 12/6/2006
2
Outline

• Environment and the problem
• Speed vs. accuracy
• Correction of chromatic distortion
• Color classification
• Blob formation
• Object detection
• Line detection
• Localization

3
(No Transcript)
4
The Problem

• Inputs
• 30 frame/sec, 208160 pictures in YCrCb format
• Chromatic distortion, noise
• Joints values
• Other sensors
• Output
• Distances and angels of robot relative to objects
• Constant objects ? Localization
• Ball, goals,

5
Overview
6
Chromatic Distortion
7
Chromatic Distortion CorrectionFormula
8
Chromatic Distortion CorrectionSample Results
9
Chromatic Distortion CorrectionImplementation

• 4 pre-computed tables
• (x, y) ? r
• (observed value of a color component Cr, r) ?
  actual value
• (observed value of a color component Cb, r) ?
  actual value
• (observed value of a color component Y, r) ?
  actual value

10
Color Classification

• Inexpensive
• 8 colors 1 unknown color
• Color look-up table
• 128128128
• HSL color space
• Basic, fundamental!

11
(No Transcript)
12
Color Learning

• A large number of samples of each color
• Learning tools
• Average point of samples of each color
• 8 points standing for 8 colors in the HSL space
• Color class of each look-up tables cell

13
Color Learning (contd)

• Similarity of each cell of LUT to each color class

14
Color Classes in YCrCb Space

• The 256256256 color space cube from which a
  192192192 cube is cut.

15
Color Classification Sample Result
16
Blob Formation

• Blob color
• Blob size
• Blob density

17
Object Detection The First Step

• A pixel in the picture ?
• A point in 3-dimensional outside world
• Camera position/angle
• Project in the direction of the pixel, crosses
  the ground
• ? The point

18
Line Detection

• Search for green-white transitions
• BFS in green-white edges
• Optimized tolerance
• Obtain lines in the outside world
• Line intersections patterns L, T
• Localization

19
Line Detection Sample Result
20
Localization

• Landmarks
• Motion feedback
• Field lines
• Goals

21
Sample Goal Detection
22
Experimental Results
23
Future Work

• Adaptable accuracy

24

• Questions?