12-08-2015 дата публикации
Номер: CN104835145A
Принадлежит:
The invention discloses a foreground detection method based on a self-adaptive Codebook background model. The method comprises: firstly, using a training sequence to train to obtain a Codebook background model, according to the Codebook background model, forming a background image, performing background subtraction on a to-be-detected image to obtain a foreground binary image, deleting the foreground region whose area is smaller than a preset threshold, updating corresponding pixel point information to the Codebook background model, setting a pixel neighborhood for each pixel point in each foreground region block, according to HOG feature vectors of regions corresponding to pixel neighborhoods in a to-be-detected image grey-scale map and a background image grey-scale map, determining whether the pixel point is a pseudo- foreground pixel point, if yes, updating corresponding pixel point information to the Codebook background model; and then, combining a brightness range and a cross-correlation ...
Подробнее