An iterative refinement method for pose estimation from planar target

In theory, the relative pose between the object and the camera can be uniquely estimated from a minimum of four coplanar but noncollinear target points. In practice, the observed image points are always perturbed by homogeneous Gaussian noise, and the pose ambiguities are easily happened. In this paper, an iterative refinement approach with a robust filter strategy is proposed. It utilizes the pose parameters obtained from non-iterative pose estimation algorithms as initialization to perform undistortion and unprojection of original images to a canonical fronto-parallel plane. This canonical plane is then used to localize the target points and recompute the pose parameters in an iterative refinement until convergence. Based on this solution, we derive a refinement algorithm which increases the accuracy of target points localization and consequently of pose estimation from a planar target. Besides, the filter strategy adopts the geometric error instead of algebraic error in the pose selection. The advantages of our method are demonstrated by through testing on both synthetic and real data.

I. Introduction

Pose estimation is indispensable to a variety of visual sensing applications, such as robot hand-eye calibration [1]–[4], vehicle pose estimation [5]–[8], and optical measurement [9]–[11]. When the intrinsic parameters of a camera and a set of correspondences between 3D points and their 2D projections are available, the six-degrees-of-freedom pose of the object in the camera coordinate system could be determined, which is also called Perspective--Point problem .