epnp an accurate o(strongnstrong)solution to the pnp problem.pdfVIP

Int J Comput Vis DOI 10.1007/s11263-008-0152-6 EPnP: An Accurate O(n) Solution to the PnP Problem Vincent Lepetit · Francesc Moreno-Noguer · Pascal Fua Received: 15 April 2008 / Accepted: 25 June 2008 © Springer Science+Business Media, LLC 2008 Abstract We propose a non-iterative solution to the PnP 1 Introduction problem—the estimation of the pose of a calibrated camera from n 3D-to-2D point correspondences—whose computa- The aim of the Perspective-n-Point problem—PnP in short— tional complexity grows linearly with n. This is in contrast is to determine the position and orientation of a camera 5 8 given its intrinsic parameters and a set of n correspondences to state-of-the-art methods that are O(n ) or even O(n ), between 3D points and their 2D projections. It has many without being more accurate. Our method is applicable for applications in Computer Vision, Robotics, Augmented Re- all n ≥ 4 and handles properly both planar and non-planar ality and has received much attention in both the Photogram- configurations. Our central idea is to express the n 3D points metry (McGlove et al. 2004) and Computer Vision (Hartley as a weighted sum of four virtual control points. The prob- and Zisserman 2000) communities. In particular, applica- lem then reduces to estimating the coordinates of these con- tions such as feature point-based camera tracking (Skrypnyk trol points in the camera referential, which can be done in and Lowe 2004; Lepetit and Fua 2006) require dealing with O(n) time by expressing these coordinates as weighted sum hundreds of noisy feature points in