0

If we have multiple image pairs and the feature correspondences from the same stereo setup , how can we leverage all of them to find the essential matrix.

The openCV cv2.findEssentialMat function takes points1 and points2 arguments that belong to one image pair.

In the hopes of identifying better inliers by the RANSAC with more feature correspondences, is there a way to incorporate the correspondences from multiple image pairs for the essential matrix calculation? And thereby the extrinsics (roll, pitch, yaw)

Right now I am calculating essential matrix for each image pair, cv2.recoverPose to get the rotation_matrix, and then averaging the roll,pitch, yaw of all the image pairs. Is there a more robust method?

Edit: Clarification: Multiple image pairs from the same stereo setup

Community
  • 1
  • 1
Sandeep Menon
  • 45
  • 5
  • the E/F matrix expresses the relationship between two cameras, not multiple cameras. if you have multiple, you also have multiple matrices. -- pictures please. I'd like some clarification on your situation. what exactly is the situation? how many cameras? mounted how, stereo? multi rig? scene and cameras, what can move, what can't move? – Christoph Rackwitz Dec 17 '21 at 10:01
  • @ChristophRackwitz There is only one stereo setup and multiple image pairs from the same stereo. – Sandeep Menon Dec 17 '21 at 13:10
  • dump them all together then. I think that should work. – Christoph Rackwitz Dec 17 '21 at 20:16

0 Answers0