image keypoints matching for registration of sparse cloud
I have a stereo camera and I want to use sparse point cloud obtained from
triangulation of SURF keypoints to relocalize it in different views of the
I use the pcl implementation of icp for registering the two clouds, and
obtain the relative positioning between the two different views. At the
moment I use a guess given by visual odometry for helping icp.
Since I have already SURF descriptors available for my points, I'd like to
perform matching and feed it to opencv to improve it's robustness.
I extract and match keypoints with opencv.
Is it possible to create from scratch some pcl data structure representing
matching and give it to icp?