Multi-mode video behavior analysis method for detecting internet violent harmful scene

The invention discloses a multi-mode video behavior analysis method for detecting internet violent harmful scenes. The method mainly comprises three stages of video scene character quick positioning detection, video scene behavior judgment and video scene harmful degree qualification. According to the method, on the basis of behavior recognition, the target person emotion and the network comment bullet screen emotion are taken as judgment dimensions, and accurate discovery of a specific harmful scene is realized. 本发明公开了检测互联网暴力有害场景的多模态视频行为分析方法，主要包括视频场景人物快速定位检测、视频场景行为判别、视频场景有害程度定性三个阶段，该方法在行为识别的基础上，将目标人物情绪、网络评论弹幕情绪作为判定维度，实现对特定有害场景的精准发现。