Selective video transmission by means of virtual reality based object extraction

This paper presents a new technique to extract objects from a real complex background so that a video sequence can be decomposed into a set of objects as required for object-oriented video compression techniques. The proposed method is based on a background subtraction technique. However, instead of using a fixed background, the system relies on predicting one from a previously constructed virtual model of the environment. Thus, camera movements are allowed. These movements are estimated by means of a tracker device. We also present the virtual model construction technique for indoor environments. The method has been successfully tested for several different video sequences including capture errors, partially mapped virtual environments and camera positioning errors. Further work will focus on extending the virtual models not only to environment, but also to objects, and integrating the method in a MPEG4 standard compression system.