Advanced image processing for future use in space systems automation

The author describes the realization of a software-based object recognition system, which is developed by MBB/ERNO in view of use in an advanced space robotics scenario for in-orbit activities. The image processing system is ready for further tests in conjunction with an industrial robot to work in a space scenario testbed on demonstration projects. To support handling tasks such as an intelligent image sensor, object recognition is a basic feature of such a system. The objects to be recognized are two-dimensional and with nonreflective but homogeneous surfaces. Standard algorithms from region oriented image processing methods have been used for the image preprocessing. The use of polar and scalar vector methods for feature extraction is shown. For classification, a neural network has been used, as well as a standard distance classificator. All algorithms have been parallelized and implemented on a transputer network