首页>
外国专利>
Violence detection framework using spatiotemporal characteristic analysis of shading image based on deep learning
Violence detection framework using spatiotemporal characteristic analysis of shading image based on deep learning
展开▼
机译:基于深度学习的阴影图像时空特征分析的暴力检测框架
展开▼
页面导航
摘要
著录项
相似文献
摘要
The present invention relates to a violence detection framework using spatiotemporal characteristic analysis of a shading image based on deep learning, capable of improving ability and accuracy of detecting violence in images. To this end, according to the present invention, the violence detection framework for detecting violence of an image by detecting a feature point of violence in an input image including image frames provided from a video camera or a video file includes: a first step of dividing a real-time input image into images per frame; a second step of excluding red (R), green (G), and blue (B) from each of the divided images per frame to extract a 2D-based Y-frame monochrome image; a third step of sequentially accumulating a plurality of extracted 2D-based Y-frame monochrome images to convert the accumulated 2D-based Y-frame monochrome images into a Y-frame monochrome image in a 3D environment; and a fourth step of extracting a frame of a uniform layer from the converted Y-frame monochrome image in the 3D environment, and performing accumulation again to perform image convolution, and deriving a desired detection scene by using a 3*3*3 filter. Accordingly, an image optimized for network lightening and a time space is created and applied to an algorithm so as to allow the feature point of violence to be continuously remembered and re-learned on a specific layer in an image convolution process, so that the ability and the accuracy of detecting the violence in the images are improved, analysis is performed regardless of a length of an analysis frame, and consecutive actions are analyzed.
展开▼