Scalable video compression scheme for tele-surveillance applications based on cast shadow detection and modelling

This paper presents a new scalable block-based video compression scheme for video-surveillance applications. It is based on the detection and the modeling of cast shadows. Each block is classified according to its content, i.e. background, foreground, cast shadow and boundary. It is assumed that a reference image representing the fixed background of the scene is available which is generally realistic in the context of video-surveillance. The bitstream is decomposed into three layers containing respectively the foreground (basic layer), shadow and background update (optional enhancement layers) information. More bitrate can be dedicated to the coding of the foreground objects, which is most relevant information. The shadows are coded using a precise model of the ambient light in the shadow area based on a level lines representation. The detected cast shadows are removed from the original images using the reference. Experimental results show that a compression gain can be obtained.