Expressway thrown object detection method based on significance foreground matching

The invention discloses an expressway thrown object detection method based on significance foreground matching. The method comprises the following steps: firstly, carrying out foreground extraction to obtain a foreground binary image fbinary and a moving foreground frame set in a video; then, foreground frames which are not on the road surface are removed based on road segmentation, foreground frames of people and vehicles are removed based on YOLO detection, non-static foreground frames are removed based on IOU static judgment, and a throwing object candidate frame is obtained; and then, inputting a current frame local image corresponding to the candidate frame b into the improved PoolNet network model to obtain a significant foreground binary image X1, performing similarity matching on X1 and a local image X2 belonging to fbinary corresponding to b, and if the X1 is matched with the local image X2 belonging to fbinary, marking the candidate frame as a throwing object target frame. According to the method, a