5 - Intégration temporelle pour une détection robuste d'objets mobiles éloignés à partir d'une caméra en mouvement

The detection of moving objects in an outdoor and disturbed scene observed by a mobile camera is a difficult problem, whos e solution requires to take into account more than two successive images to achieve robustness . In this paper we present two ways to carry out this temporal integration in the motion detection process : first, the use of long temporal observations for the decisio n criteria, secondly the implementation of a prediction module . These observations are usually get from temporal filtering at differen t temporal scales . Prediction of mobile objects is iteratively updated using current detection results, but also controls the detectio n process of the following instant . When using a static sensor, we have only to consider successive images, but if the sensor itself i s moving, we need previously to compensate the apparent motion of the image . The methods that we have developped are teste d on various real images, and their interest is evaluated .