Video face privacy method based on optical flow

The invention provides a video face privacy method based on optical flow. The method comprises the following steps: S1, intercepting two continuous video frames in a real-time monitoring video; S2, inputting two continuous video frames into a face detection network, and detecting faces in the video frames by using a pre-trained optical flow network model and a face detection model in the face detection network; S3, detecting and outputting face position information, and dividing a face area of a face position area into n * n sub-blocks; and S4, obtaining the value of the first pixel point in the n * n sub-blocks, resetting all the pixel values in the sub-blocks as the value of the first pixel point, and presenting a mosaic effect to shield the human face. Information in the time dimension of video data is fully mined, missing detection and false detection of a face detector are reduced, face privacy is enhanced, and other picture information is not shielded while face information is not leaked. 本发明提供一种基于光流的视频