Video description method and system based on hierarchical modularization

The invention discloses a video description method based on hierarchical modularization. The method comprises the following steps: step 1, obtaining important object features in a video; step 2, according to the initial dynamic features of the video, combining important object features to obtain video action characterization; 3, video apparent characteristics are obtained, and global video representation is obtained in combination with important object characteristics and video action representation; and step 4, obtaining a final description statement of the video. According to the video description method and system based on hierarchical modularization, the corresponding relation between video information and language semantics can be fully utilized, correspondence of overall semantics and detail semantics is considered, and the video task description accuracy is remarkably improved. 本发明公开了一种基于层级模块化的视频描述方法，所述方法包括以下步骤：步骤1，获得视频中的重要物体特征；步骤2，根据视频的初始动态特征，结合重要物体特征，获得视频动作表征；步骤3，获得视频表观特征，结合重要物体特征、视频动作表征，获得全局的视频表征；步骤