I have sequences of images I am using for cv2.triangulatePoints.
From external odometry system, I already have decently accurate extrinsics of the camera.
If I didn't have extrinsic, I would find an essential matrix and cv2.recoverPose, which results in really clean triangulated points because I can use masks to filter matched points before triangulation.
However, when I extract ORB features and match using ratio test, and then triangulatePoints.
Visually inspecting, I would most often get ~80% well triangulated points but 20% noise.
On some frames, it is more 50% good points and 50% bad noise.
Is there a way to clean up these points before triangulating them? Thanks!