Cooperative object recognition

This paper introduces an approach to recognize objects in a group of observ ers. The fundamental idea of this approach is based on the usage of coopera tion at object level in order to overcome limitations of typical recognitio n applications which often lead to ambiguous interpretations. By integratin g individual hypotheses which were calculated at spatial distributed viewpo ints, the robustness of the recognition results can be increased significan tly. Experiments that compare cooperative and individual calculated results confirm the benefit of cooperation in recognition tasks. A qualitative met hod for appearance-based object recognition is used to build hypotheses for individual recognition. Based on an input image which mainly contains the target object., a translation and scale invariant representation is built. From this, a two-dimensional distribution function is generated and the sim ilarity between any two objects is determined by using non-parametric stati stical tests. The fusion of distributed information is done by the use of b ayesian networks. Such networks enable both a continuous individual and a c ooperative object recognition. (C) 2001 Published by Elsevier Science B.V.