Method and system for robust scene modelling in an image sequence

摘要

Disclosed is a method of storing a scene model (230) used for foreground/background separation of a scene (223) captured by a camera (100), the scene model comprising a plurality of visual element models (240) each associated with a set (250) of mode models (260, 270, . . . ), said method comprising determining (330) if a change in the scene satisfies a pre-determined threshold (MFGT), creating (710) a background image based on a plurality of the mode models classified as background, matching (360) an input image (210) captured by the camera and the background image by determining a similarity score, and updating the set of mode models if the similarity score satisfies a pre-determined similarity threshold by creating at least one mode model based on the received input image, said mode model being classified as background.