I have a set of segments (each segment is in a std::vector of 2 cv::Point) and if I print them in a image they're arranged in this way:
Most of them are in the two pseudo-parallel lines, other segments are noise. What I want to do is detect the two main lines and build the matching quadrilateral like this:
What is the best approach to solve this problem?
Thanks!
I guess that you want to find a rectangle that was deformed by a perspective transformation. Is this right? In this case I'd try to merge the short lines, find the four endpoints of the merged lines and use getPerspectiveTransformation to estimate the transformation. If you make a mistake during merging, the reprojection error of the transformation will be rather high so you can try to adjust your merging (maybe ignore some of the shorter lines or something similar) until you get a transformation with a good reprojection error. This approach is called RANSAC and could probably solve this problem.
Asked: 2014-09-24 11:42:01 -0600
Seen: 862 times
Last updated: Sep 25 '14
