Eye-Tracking System: Estimate Homography to do 2D to 3D integration (8)

I. Collecting other 2 datasets with circle color marker:

II. Do thresholding:

1. Can use the function: im2bw, graythresh, bwconncomp / bwlabel / lab2rgb
- im2bw
- graythresh
- bwlabel, bwconncomp, label2rgb, regionprops

2. Find the centroid
- solution 1: using 'regionprops()' get centroid (... the '+' mark in purple)
- solution 2: using a source code 'circfit()' providing in George's homework (... the '*' mark in black)

Kinect frame: after bwconncomp()

Kinect frame: after bwconncomp() and zoom-in the green marker

Hand-held image: after bwconncomp()

Hand-held frame: after bwconncomp() and zoom-in the green marker

Moreover, the Yellow Marker provides unstable segmentation results across our datasets comparing to other three:

4. Considering the upper 2 issues: 'the location of central is influenced by the distinguishable boundary of each marker' and 'Unstable segmentation result with Yellow color', so I decide to re-design my marker. Moreover, keep the marker be used in a simple way - all in the same color 'red' as following.

New Color Markers

The yellow '+' in the bottom-right image shows the location of center of the red region

* This is a preliminary simple test which is for proving that the additional black background around the red color marker can either enhance the visibility of the color marker or limit the region of this marker precisely.

* Based on observation, the color 'red' is more stable under illumination change (...yet, we don't care about illumination change; this is just kind of factor which influences our experiment)

* Because the color marker will be substitute into feature points, just keep the color marker as simple as possible. All of these 4 markers in the same red color means that we all need to worry about one value of thresholding per image instead of 4 (i.e., red, green, blue, yellow).

III. Need to use the parameter of 'projection' in maketform() instead of 'affine', because...:

Then, we are able to assign 4 pairs of points to define the matrices of transformation based on the 4 color markers. Here is the results based on this alteration: