The invention provides a dual-mode masked man event automatic detection method based on video characteristic statistics. The dual-mode masked man event automatic detection method comprises the steps of: zooming a video frame and converting the video frame into a grayscale image after acquiring a source video image, acquiring a maximum object motion foreground rectangular frame, positioning a head region coarsely, acquiring face position information, then estimating a mouth region within the obtained face range, cutting the mouth region to narrow a range, calculating a gradient map, primarily judging whether a person is a masked man through calculating number of connected domains on the basis of the obtained gradient map, and calculating proportion of the areas of the top two connected domains for further judging whether the person is a masked man. The dual-mode masked man event automatic detection method based on video characteristic statistics does not have specific requirements for a video scene, is simple and efficient in the calculation method, can perform real-time processing, can be applied to real-time investigation and case solving, saves suspicious masked man information, and can provide first-hand information for evidence collection after a crime is committed.