This paper discusses the problem of high performance obstacle recognition. The obstacle recognition is a critical task for mobile robots with autonomous driving possibility. The problem setting, applicable requirements for method and sensory system, overview of existing approaches are described in this paper. The paper also presents the analysis of road boundaries detection methods, surface approximation model definition, geometric primitives and characteristic tags detection techniques, neural networks. The 3D range-finder-based method of structural obstacle recognition is suggested including descriptions of testing technique and quality parameters estimation. The functional prototype of obstacle recognition system which uses the proposed method is also described.