Pose Estimation of Two-Pose 3D Models Using the Base Tangent Plane and Stability Constraints

A pose estimation technique is presented to determine coarse registration parameters between two 3D models of an object. The models are reconstructed by merging multi-view range images of two different poses of the object, which is arbitrarily placed on a turntable. The result of pose estimation will facilitate registration refinement and integration of two 3D models. Registration refinement of two 3D models usually requires a rough estimate of pose between them. However, obtaining such estimate is a difficult problem and computationally expensive. We introduce a simple pose estimation technique based on matching tangent planes of a 3D model with the base tangent plane (BTP) which is invariant for a vision system. In order to reduce computation time, we employ geometric constraints to find consistent and stable tangent planes (STP). A STP is a plane on which the object can rest in a stable state on a turntable. By matching the base tangent plane of one 3D model to a stable tangent plane of the other model, we derive a pose transformation matrix and register the models to a common coordinate system. We find the best-matching pose transformation matrix which minimizes a pose error beween two models. Experimental results on several real objects are presented to demonstrate the effectiveness of the method.