Selection of image features for robot vision system

In robot visual servo the system performance has a close relationship with image features. This paper mainly discusses two kinds of image features. One is geometrical moments, and the other is eigen-space transform features, which can be regarded as a kind of overall image features. Using overall features may avoid designing artificial marks on the objects, and make image feature extraction algorithms easy be used for different objects. The overall image features can also avoid the problems of limited robot manipulating scope and inconvenience in computation. The eigen-space transform algorithm is detailedly discussed. The experiment results show that the moment features are sensitive to disturbance from the environment, and the features from eigen-space transform are more anti-noise than the moment features. We successfully use this kind of image features in robot visual servo and in robot motion simulation.