Developing rigid constraint for the estimation of pose and structure from a single image

Abstract

Pose and structure estimation from a single image is a fundamental problem in machine vision and multiple sensor fusion and integration. In this paper we propose using rigid constraints described in different coordinate frames to iteratively estimate structural and camera pose parameters. Using geometric properties of reflected correspondences we put forward a new concept, the reflected pole of a rigid transformation. The reflected pole represents a general analysis of transformations that can be applied to both 2D and 3D transformations. We demonstrate how the concept is applied to calibration by proposing an iterative method to estimate the structural parameters of objects. The method is based on a coarse-to-fine strategy in which initial estimation is obtained through a classical linear algorithm which is then refined by iteration. For a comparative study of performance, we also implemented an extended motion estimation algorithm (from 2D-2D to 3D-2D case) based on epipolar geometry.