Object identification based on uncertain sparse sensor information

Abstract

Listen

Translate article

This paper discusses the problem of object identification based on spatially sparse uncertain information. First, it is assumed that the sensor information is composed of two kinds of information, i.e., the geometrical parameter distribution and the decision tree. The description format for the uncertain sensor information is defined. Then a method is presented where the set of uncertain sensor information obtained by the measurement of the object is combined (fusion), and the match between the object and the model is examined. Finally, the case of measurement of the two-dimensional figure by the sparse shape sensor is considered. The process of fusion is demonstrated and the usefulness of the proposed method is shown by a matching experiment. In the proposed method, an object identification procedure is employed where the information obtained from several sensors is described as geometrical parameters together with their error distributions, and the match to the model is examined by combining those parameters. The information, for which more than one decision tree interpretation is possible, is described as the basic probability of the Dempster-Shafer theory. The parameters are fused for each interpretation, and the reliability of the result of fusion is estimated by the combination rule. As a simulation experiment, the case with the tactile sensor in contact with a column object is considered, and the object identification process is demonstrated.