Development of a scene recognition system with imprecise descriptions

This paper presents the development of a scene interpretation system with imprecise descriptions in a 3D robotics context. The problem of "scene recognition" is presented as the matching of images with a priori descriptions given in a high level language. The mission consists in the identification of the position of an object in a scene with a vision system situated at the end of a manipulator. The position of the object is given by the human operator in a pseudo-natural language and the environment is only known through the fuzzy hints of the mission description. The organization of the system and the particular aspects of the scene recognition method are presented. We conclude with the first results of the complete system.