A Hybrid System to Detect Hand Orientation in Stereo Images

The recognition of hands and the 3-dimensional characterization of hand orientation is a diicult and practically important vision problem. In this work we propose a hybrid system attaching artiicial neural networks to concepts of a semantic network to solve this problem. The neural networks are attached as a holistic representation at varying levels of the abstraction hierarchy modeled by the semantic network. In this way we are able to combine the advantages of both techniques. However, it can also be applied to other problems in computer vision and tasks of signal interpretation in general.