Project 3: Camera Calibration and Fundamental Matrix Estimation with RANSAC
Go back to Home
Results:
I used VLFeat(which uses SIFT) to extract matching points. To select points, I ran my RANSAC, both with the non normalized fundamental matrix estimator and with the normalized fundamental matrix estimator. For visualization purposes, I sorted the matching points by their distance from the inteded point. I displayed the top 30 points.
Default values used: Threshold = 0.05, Proportion = 0.8, Iterations = 3000
Format of the image tables:
Matching points and epipolar lines superimposed on original images, again determined with non normalized fundamental estimator
Corresponding points in the images determined with non normalized fundamental estimator.
Matching points and epipolar lines superimposed on original images, again determined with non normalized fundamental estimator
Corresponding points in the images determined with non normalized fundamental estimator.
Notre Dame
Result:Both cases work well. Notice the intersection at a point of the epipolar lines in the normalized version.
Mount Rushmore<
Result: Here, the normalized version seems to be performing slightly worse than the non normalized version. Probably because of the randomness involved.
Gaudi
Result: Here, the normalized version clearly outperforms the non normalized version.
Woodruff<
Result: Here too, the normalized version clearly outperforms the non normalized version.
Full res images here
Go back to Home