Rotation scan v2
Previous version used a blob detector to detect the keypoint of a corner marker. This was good enough for scan rotation, however for scan shifting this is not sufficient. Therefore, it was necessary to implement a better detector for corner marker keypoints.
This is now done roughly as follows:
- Use the blob detector as previously
- Create an image descriptor/patch which is basically a crop around the keypoint
- Use a corner detector on the image patch to gain a better keypoint
P.S. While the branch is named rotation-scan-harris, harris corner detector is not used. There was in the end an even simpler and better solution for now which uses cv2.goodFeaturesToTrack()
Edited by Nick Cleintuar