Improvement of the Human Action Recognition Algorithm by the Pre-processing of Input Data

Abstract The paper presents an approach to recognizing human actions using additional preprocessing stage of input data. growing volumes video information do not always allow support the quality data at a high level; this can cause limitations in further processing digital In regard, it becomes urgent introduce image enhancement into algorithm for video. proposed method includes three main steps: enhancement, constructing descriptor, and classification. presented is based on combined local global frequency domain. basic idea alfa-rooting apply different disjoint blocks with sizes. To solve problem three-dimensional microblock dense difference (3D DMD) used, which provides highly oriented representation regions by tightly capturing microblocks within each region several orientations scales. 3D DMD has advantages over other methods: higher efficiency compared existing methods; minimal computational costs when integrated image; low dimension; ease implementation; does require settings. modification allows increase productivity 2-4%.