Homework #1 : Panorama from multiple photographs

 

 
 
 
 The two images used for input with correspondence points.

 
 
Gold standard with 4 points.

 
Gold standard with 10 points.

 
Gold standard with 20 points.

 
Gold standard with 20 points and bilinear alpha falloff from image edges.

 
 
Errors
4 points 10 points 20 points
DLT algebraic error 0 0.0004099 0.0008163
symmetric error 0 29.8661 56.5824
Normalized DLT algebraic error 0 0.003945 0.0079
symmetric error 0 29.6419 55.9887
Gold Standard Sampson error 0 7.2768 13.6052
symmetric error 0 29.5580 55.9887

 
Errors (normalized)
4 points 10 points 20 points
DLT algebraic error 0 0.00004099 0.00004082
symmetric error 0 2.9866 2.8291
Normalized DLT algebraic error 0 0.0003945 0.0003971
symmetric error 0 2.9642 2.7994
Gold Standard Sampson error 0 .7277 0.6803
symmetric error 0 2.9558 2.7994




The normalized error table divides the total error by the number of points. Here we can see that increasing the number of points decreases the error. It is interesting that the normalized DLT yields a larger algebraic error, but the geometric error is less. That doesn't seem right.
 

 Data (images, points) (91K)
 Code (4K)