Faculty of Informatics Vienna University of Technology Institute of Computer Aided Automation PRIP Home PRIP Home
Personal tools
You are here: Home People Julian Stöttinger FeEval Homography Scale and Rotation

Homography Scale and Rotation

Homography estimation for challenge "scale and rotation". Matlab function source download.

For evaluation of the FeEval challenge scale and rotation, affine 2d homography has to be known.

To evaluate to robustness to scale and rotation we increasingly shrink the videos to a final size of 30% of the original size and rotate them by 10% for every level.

The MATLAB function can be found here.
The Homography matrices are available here.

1-scalerot-4.jpg

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

The homography matrices are estimated as follows:

identity matrix
1 0 0
0 1 0
0 0  1

 

translation matrix
1 0 x
0 1 y
0 0  1

 

rotation matrix
cos(angle) sin(angle) 0
-sin(angle) cos(angle) 0
0 0 1

scaling matrix
scaling factor 0 0
0 scaling factor 0
0 0 1

 The final homography matrix is then computed by

H = inv(H_trans) * H_scale * H_rot * H_trans

H_trans is the translation to the frame center.

Example for the TV show videos:

H1 = estimate_homography(312, 163, -10, 0.9)
H2 = estimate_homography(312, 163, -20, 0.8)
H3 = estimate_homography(312, 163, -30, 0.7)
H4 = estimate_homography(312, 163, -40, 0.6)
H5 = estimate_homography(312, 163, -50, 0.5)
H6 = estimate_homography(312, 163, -60, 0.4)
H7 = estimate_homography(312, 163, -70, 0.3)