Development of an object recognition and tracking for anthropomorphic robots using a stereo vision system

Abstract

The DLR Hand-Arm-System is the first component of a new humanoid robot which is able to perform in a wide range of applications. For na\-vi\-ga\-tion and interaction applications reliable and safely information of the surrounding environment are required. Optical sensors in combination with an object recognition system are commonly used to realize this task. The objective of this thesis is to improve the recognition rate of such systems by allowing the user to choose an appropriate recognition method for a given object from a collection of several possibilities. Therefore, a vision system consisting of a Kinect camera and an adaptable object recognition software framework, containing several object recognition approaches, is realized. Dependent on the search object and the surrounding area a suitable combination of different methods can be selected to detect and localize an object robustly. The object pose, resulting from the localization, serves in a subsequent step as input for a grasping application of the Hand-Arm-System.