What distance should we calibrate at and how much does this distance affect our results?

PART 1

NOTE: Data collection for this project can be done jointly with the WalkingExperiment

Instructions

Set up 9 calibration points on two different walls so that they fill the scene camera nicely from 5m away (wall 1) and from 10m away (wall 2). Place 4 additional points within calibration points and outside of calibration points. Total 17 points per wall. Run subjects (more detailed instructions for running subjects on WalkingExperiment):

1) Have subject look at center calibration point and wiggle headgear (use these data to get camera gains).

2) Have subject look at all 9 points on wall 1 (use these data to get eye gains).

3) Have subject look at "additional" points on wall 1.

4) Have subject turn and look at all 9 points on wall 2.

5) Have subject look at "additional" points on wall 2.

After data has been collected:

6) Run tapes through ISCAN or Yarbus to get Pupil and CR values.

7) Digitize tapes.

8) Get frame numbers for when subject is looking at each point.

9) Open each of these frames of the scene videos and get coordinates of the points the subject is looking at (during that frame).

Subjects run

• (example entry) Subject 1 - June 1, 2008
  • subject wore contact lenses
  • other notes that may be relevant, etc.

Calibrated

• (example) Subjects: 1, 2, 3 (attach data files to this wiki page!)
• (example) Issues: Subject 4 could not be calibrated because...

Gain values (doesn't have to be done this summer)

• (example entry) Subject 1
  • Eye Gain = .52 H, .37 V
  • Cam Gain = .89 H, .85 V

PART 2

Prepare data

Create "mapping matrix" in MATLAB using code that Sue started, Pupil and CR coordinates for marked frames and corresponding selected scene coordinates USING:

1) the top left and bottom right points from wall 1. Apply to all data.

2) the corner points and center point from wall 1. Apply to all data.

3) all 9 points from wall 1. Apply to all data.

4) Repeat 1-3 for wall 2.

... maybe try different methods for creating matrix also - affine, projective, etc. MATLAB lets you change these settings but some methods require more control points than others.

Analyze data

• Calculate errors for calibration points (here JUST use the points used to make the matrices, not all 9 points per se) as true coordinates (as manually selected) compared to POR coordinates from application of matrix. (2-9 points)

• Calculate error for non-calibration points. (25-32 points)

• Questions to answer: (update with more)
  • Is 9 points better than 5?
  • How bad is it to just use 2 calibration points?
  • How bad is parallax (for both distances, calibrating further from test and calibrating closer than test)
  • Compare the different methods (affine, projective, etc.)

Page permissions

• Set ALLOWTOPICCHANGE = VplGroup

-- SueKola - 15 May 2008