Video watermark embedding method and device based on multi-source information, equipment and medium

Abstract

The application provides a video watermark embedding method and device based on multi-source information, equipment and medium. The method comprises the following steps: acquiring an original video segment collected by a camera terminal in real time; determining a visual style based on each frame of image to be rendered in the original video segment; generating a motion residual track according to first skeleton key points of adjacent frame images in the original video segment; performing synchronous synthesis processing on the visual style and the motion residual track to obtain a dynamic watermark layer; acquiring a synchronous physiological metadata stream synchronized with the original video segment; performing frame-level composite embedding of the dynamic watermark layer and the synchronous physiological metadata stream into original video frames of the original video segment to obtain a target video segment carrying biomechanical identification features; and deeply fusing the action track and the visual style with the physiological metadata stream, which greatly improves the efficiency of the biomechanical identification features, realizes the "what you see is what you get" analysis experience, and ensures the synchronization of the data stream and the video stream at a millisecond level.

Images