PROBLEM TO BE SOLVED: To provide a frameworking method and apparatus for detecting violence from video. A violence detection frameworking method includes a first step, 2D, of extracting a two-dimensional (2D) -based luminance component Y image by removing color difference components U and V from one frame image included in an input image. The second stage and the 3D-based Y image group in which the base Y-images are sequentially accumulated in 3D (3D), and only the equally spaced frames are extracted to obtain a 3D (3D) -based Y image group. By including a third step of deriving a scene of violence detection by performing a video convolution using a 3 * 3 * 3 filter, the weight of the network and the time-space optimized video are included. It is made and applied to the algorithm, and the characteristic points of violence are continuously memorized and re-learned in a specific layer during the image convolution process. [Selection diagram] FIG.
展开▼
