Image-based ponding trapped person risk detection method and device, and medium

Abstract

The invention discloses an image-based accumulated water trapped person risk detection method and device and a medium, belongs to the technical field of machine vision, and is used for solving the technical problems that an existing risk early warning mechanism for trapped persons in water is not comprehensive enough, trapped risk early warning is difficult to carry out, and the continuous trapped duration of the trapped persons is difficult to calculate. The method comprises the following steps: carrying out pixel-by-pixel segmentation processing on a ponding area on a video frame in a ponding risk area, and determining a ponding disaster image; performing waterlogging trapped identification on pedestrian targets in the waterlogging disaster image to obtain trapped candidate targets; pixel distance-based ponding line height comparison is carried out on the human body key points in the trapped candidate targets, and trapped person information is determined; extracting a target feature vector of the trapped person information; and performing cosine similarity calculation on the historical feature vector in the historical frame and the current feature vector in the current frame, and integrating the trapped time of the person and the trapped person information to obtain the risk early warning information of the trapped person.

Images