The invention relates to a video recognition based river tidal bore detection method. In the method, tidal bore video information is acquired through video detection devices which are arranged along a river and is transmitted to a tidal bore monitoring center through a wireless network after subjected to analytical processing and recognition; and a tidal bore safety forecast is sent to each danger point on the upstream side of the river by the center according to the tidal bore information. A real-time video signal of the surface of the river is acquired through the video detection device; a video frame is processed; a mixture Gaussian background model is established; a background image is constructed; a moving object target is extracted by comparing a current frame of image with the background image while the background image is updated by using the Gaussian model, image elements of a moving object are calculated; and the existence of the tidal bore at a monitoring point is automatically recognized according to a preset threshold and a tidal bore image is transmitted to the monitoring center. By adopting the method, remote monitoring can be realized and safety of monitoring can be improved.