Perspective-n-Point

Recovering a camera’s pose from a set of known 3D points and where each one lands in the image.

PnP is a classic computer vision problem: given the camera’s intrinsic parameters, n 3D points, and their 2D projections in the image, find the camera’s rotation and translation relative to those points — 6 degrees of freedom in total. At minimum, 3 point pairs are needed (P3P), but 3 points can yield up to 4 solutions, so a 4th point is usually needed to disambiguate. EPnP, proposed by Lepetit and colleagues in 2009, solves the problem using 4 virtual control points, with computation that scales linearly in the number of points. Real matches often include errors, so PnP is typically paired with Random Sample Consensus (RANSAC) to reject outliers; OpenCV’s solvePnP and solvePnPRansac are the most commonly used implementations. In robotics, it is often used to recover a pose from the corners of a calibration board or an ArUco marker.

ExampleA camera sees an ArUco marker of known side length; feeding the 3D coordinates of its 4 corners and their detected pixel positions into solvePnP gives the marker’s position and orientation relative to the camera.

Also called

PnP, PnP Problem, solvePnP

Related

Camera Intrinsics · 6D Object Pose Estimation · Random Sample Consensus · ArUco Marker · Hand-Eye Calibration · OpenCV (Open Source Computer Vision Library)

Sources

Perspective-n-Point - Wikipedia