0

I am using OpenCV for some project. I want to use part of a planar appearance to track this planar with template matching. Is there some method to choose stabler patches for template matching? Any tips would be appreciated. Thanks in advance!

flyzhao
  • 348
  • 4
  • 15

1 Answers1

2

One standard method is to detect strong keypoints and extract regions around them. If the keypoints are stable, these regions will also be stable. In fact, you may be better off extracting keypoints and then their descriptors for matching, because these will be invariant to scale and rotation (assuming you use SURF or SIFT). Take a look at this example in OpenCV.

Zaphod
  • 1,927
  • 11
  • 13
  • Thank you for your reply. Currently i am uring SURF, but I don't want use surf each frame, because of time consuming. So I want to use template matching between consequential frames. It could reduce running time ,if we can know which part of the image is stable for template matching. – flyzhao May 17 '13 at 09:17
  • I'm fairly sure that template matching will end up slower than SURF matching, to be honest. But definitely worth trying. You could identify regions for matching by detecting SURF keypoints in the first frame, followed by template matching of those regions in the following frames. – Zaphod May 17 '13 at 09:48
  • I agree with you ,if i try to do template matching along the whole image, it would be slower. I detected SURF keypoints or ORB with the first frame to identify regions,then, i just need use template matching to get the transfer matrix between two frames, so only need to do a few times with template matching. I try optical flow, but is not stable as i expected. Thank you~ – flyzhao May 17 '13 at 11:11