In this paper, we focus on the problem of foreground segmentation in outdoor environment with a static TV camera. Our application context is the visual surveillance of archeological sites. In this context the main aim is to detect the presence of people and to recognize their gestures in order to individuate the illegal actions. In this paper we concentrate solely on the primary step of moving object detection. In particular, the system should be able of recovering the true shape of the moving objects in order to allow to a classifier to discriminate a people from any other moving object as car and animals. Moreover, the system should not be sensitive to changes in lighting, weather, number of people, etc., and it is required to work autonomously for long periods of time. A main problem in analyzing real outdoor daylight scenes is to deal with shadows cast by moving objects such as vehicles or pedestrians. Cast shadows are often detected as a part of the moving objects since they move in the same way. When the detected objects contain shadows, large errors may occur with respect their recognition. In this paper a new approach for cast shadow removing is proposed. Our idea is to detect shadow points as points that are static for a short temporal sequence and that are characterized by a photometric gain, with respect the reference background image, that is lower than unit and that we estimate for each new image through an optimization approach.