3D Pose Estimation for Robot Mikado

There is a persistent need to establish a spatial mapping between two or more 2D or 3D representations of an object. This task is very common in the field of science, medicine, engineering, as well as in the field of robotics. This spatial mapping is called object registration or object pose (position and orientation) estimation and its goal is to determine a spatial transformation that best aligns two object descriptions. This thesis work focuses on the implementation and evaluation of methods for free-form object registration from 3D point clouds. Enhancement to the registration accuracy is achieved by using more informative features (color) to existing methods (ICP, PPF). This thesis, for example, describes a system for accurate object registration from a 3D point cloud which is vital in the task of robotic grasping. 3D pose estimation experiments demonstrate that proper down-sampling and the use of color information in our setup significantly increase registration accuracy. Besides having multiple areas in which object registration can be applied, the chosen field of application for this thesis is oriented to a future robotic grasping task.