Multisensor background extraction and updating for moving target detection

This paper proposes a new method for multisensor background extraction and updating aimed at surveillance and target detection applications. The background scene extraction is based on robust multisensor change detection of moving objects in the scene. An iterative mechanism updates the background estimate using this information thereby ignoring transient objects but allowing for slow changes in scene illumination. A new multisensor confidence framework is proposed that combines change information from all available sensors to greatly improve the reliability of true scene change detection therefore the reliability of the background image estimate. The proposed method is demonstrated on several sets of registered multisensor images.