CV Triangulate Points (Two-View)

Reconstructs sparse 3D points from two views given the matched 2D points, the shared camera matrix and the relative pose (cv2.triangulatePoints) - the sparse structure half of two-view structure-from-motion. It builds the projection matrices P1 = K[I|0] (view 1) and P2 = K[R|t] (view 2, from 'CV Recover Pose'), triangulates each correspondence and divides out the homogeneous coordinate. Points are in VIEW-1 camera coordinates and only up to the same global scale as the unit translation. The 'valid' mask flags points that pass the cheirality check (positive depth in BOTH cameras) and are finite - outlier matches and points near infinity are marked 0; preview points_3d with 'Inspect CV Data' and check reproj_error (mean pixels over valid points; under ~1-2 px is a clean reconstruction). Failure-tolerant: 0 points or a degenerate pose returns empty arrays, error 0 and count 0 instead of raising.