Stereo-Based Object Recognition and Pose Estimation System

Abstract

In this chapter, the developed system for the recognition and pose estimation of objects is presented. The system consists of two subsystems: one for objects that can be segmented globally and are defined by their shape only, and the other for textured objects, which can be recognized on the basis of point features. The relation to state-of-the-art systems will be explained at the beginning of the Sections 6.1 and 6.2, respectively.

In both subsystems, a calibrated stereo system is used for calculating the object pose, leading to a more robust estimation and a higher accuracy, as will be shown in the Sections 9.1 and 9.2. Both subsystems offer two operation modes: one for a complete scene analysis, and the other for the recognition and pose estimation for a specific object representation. The latter mode allows fast tracking of a single object, which is in particular necessary for visual servoing applications.