State of the Art in Object Recognition and Pose Estimation

Abstract

Computer vision is a vast research area with many applications, both in industry and artificial intelligence. One of the most traditional disciplines is the recognition and pose estimation of objects. In the following, an overview of existing methods and algorithms that are relevant and useful for humanoid robots is given.

A humanoid robot system designed for grasping of objects in a real-world scenario sets the highest requirements to object recognition and pose estimation, more than any other application. Not only have the computations to be performed in reasonable time and objects have to be recognized in an arbitrary scene, but the pose estimation algorithms have also to provide full 6-DoF pose information with sufficient accuracy. In the following, existing approaches to object recognition and pose estimation are presented, explaining the pros and cons of each for the desired application. Since object recognition on its own is already a huge research area, even a whole book dealing with the state of the art only would hardly offer enough space for giving a complete overview. Therefore, only those approaches are presented that are related to pose estimation as well and are relevant in the context of humanoid robots.