Queue based fast background modelling and fast hysteresis thresholding for better foreground segmentation

Abstract

Background subtraction is a real time effective technique for detecting moving foreground objects in image sequences from a static camera. Background modelling plays an important role in this technique of foreground object detection. Active real time background modelling in presence of moving foreground objects in the scene and adaption of background model to gradual changes due to gradual illumination changes and addition of new immoveable objects into the scene are addressed in this paper. We present a queue based algorithm for real time, active, and adaptive background modelling. Segmentation of the foreground and robust detection of shadow is performed via comparison with background statistics in YCrCb color space. The problem of a single foreground object splitting into two or more segments due to similarity of foreground pixel color with the background in most cases can be ameliorated with the use of a fast single pass the hysteresis thresholding technique. We demonstrate various results of background modelling, segmentation and shadow detection results for both indoor and outdoor scenes.