The aim of this paper is to describe a flexible and
robust background management algorithm. In these years various techniques were proposed to segment the images from a video
stream sequence, and detect interesting dynamic objects. Many works faced the problem to segment the image in indoor
environment for human detection and intelligent room applications.
In these works, both accuracy and efficiency depend on the
background model they used. Specific high performances models suffer of some limitations in chaotic unstructured environments.
Long video stream sequences changes in light condition, and object and human displacement. In those environments, dynamic to stable objects and humans can be absorbed in the background, and then
become invisible to the system. In this work we propose an approach to combine low level and high level information to improve the background management and to solve unpredictable object dynamic problems. Experimental recall and precision results show improved performances with respect to popular background management algorithms. Finally, a real application is shown and discussed.
