This post is a stub for now, but I will be updating it with more information tonight or tomorrow, so check back on it.
I just wanted to show some quick results from a new version of the SIFT image matcher I’ve built.
The new OpenFrameworks project is on my GitHub as
SIFT_filterMatches_homography
This version of the project includes:
- keypoint match filtering, to separate the bad keypoints (outliers) from the good matches
- 3D transform (using homography) of the “query” image (what we’re looking for) onto where it is in the “train” image (what were looking in).
Right now, a simple trapezoid is drawn, but the next step will be warping the query image. - SIFTMatcher class contains SIFT detection, description, keypoint matching, match filtering, and homography transform functions
Exciting results! Actually, this is the first step towards DIY augmented reality if you’re so inclined.
Sample Results:
Thanks to Bryan Ma for providing the soundtrack to today’s project: