Hrytsyk V. Models and Tools for Adaptive Processing Videostreams in Computer Vision Systems

Goal of the research is to improve the accuracy of object recognition in the complexity field of attention in real time by developing computer models of perception, methods and architectures of adaptive processing of video streams in computer vision systems aimed at intelligent data processing and its parallelization. The subject of the research is models and tools for adaptive processing of video streams in computer vision systems. Main results of the research are the following: - bad formalized pattern recognition method is proposed, which is based on non-similar areas on complicated fields; - classes of algorithms method for tasks of computer vision are showed. Those classes include parallel, recursive and parallel-recursive technologies of data processing. When system use this method this system is optimized to realization of special tasks. - hardware oriented method is paralleling video stream, this method based on thermal represented of algorithms. This representation realizes parallel processing of video stream down to bit level. - computer vision systems architectures synthesis method is proposed. It is based on on a space-time representation of algorithms. This representation let the system processing video streams in real time with minimal hardware resources. Computer Vision systems creation theory has next development: - pattern recognition mechanism, adaptive video input systems, real time processing systems theory; - operational changing resolution on non-integral level method by using software-controlled raster; it realized resolution of input system up to 2-3 times. - methodology for evaluating quality, effectiveness and reliability of data transmission in computer vision systems.