solvePnP

solvePnP is a function in OpenCV that solves the Perspective-n-Point (PnP) problem to estimate the pose of a calibrated camera from a set of 3D–2D correspondences. Given a set of 3D object points in the object coordinate system and their corresponding 2D image points, the function computes the rotation and translation that align the 3D points with their image projections under the given camera intrinsics.

Inputs include objectPoints (3D points), imagePoints (2D points in pixels), cameraMatrix (intrinsic matrix), and distCoeffs (distortion coefficients). The outputs are rvec and tvec, where rvec is the Rodrigues rotation vector and tvec is the translation vector describing the camera pose relative to the object (or equivalently the object pose in the camera frame, depending on convention). A flags parameter selects the algorithm used to solve the PnP problem, with options such as iterative, EPNP, P3P, AP3P, DLS, UPNP, and IPPE variants. The iterative method is the default and typically robust when many points are available.

In addition to solvePnP, OpenCV provides solvePnPRansac for robust pose estimation in the presence of outliers, combining PnP with RANSAC to reject inconsistent correspondences. The choice of algorithm balances accuracy, speed, and the number of required points; for example, P3P variants require at least three points, while others can operate with larger sets.

Common applications include augmented reality, robotic navigation, and pose estimation of known objects in a scene. Practical use requires accurate camera calibration and well-distributed point correspondences to avoid degenerate configurations.