Pose Estimation of Free-form Objects (Theory and Experiments)

Abstract

In this report we present geometric foundations and an algorithmic approach to deal with the 2D-3D pose estimation problem for free-form surface models. This work is an extension to our former works presented in [28]. But whereas in [28] 1D contour models are treated, here the extension to 2D free-form surface models is presented. Similar to our former works we use a parametric representation of surfaces and apply Fourier transformations to gain low-pass descriptions of the object. We present an algorithm for pose estimation, which uses the silhouette of the object as image information and recovers the 3D pose of the object even for changing aspects of the object during image sequences. We further present extensions to couple surface and contour information on objects and show the potential of our chosen approach even for complex objects and scenes.