The Advanced Visual Monitoring Project at IRST

11 Final Remarks Thank to fruitful collaborations established with nal users and implementors, a substantial portion of our activity is devoted to transferring the technologies developed within the project into advanced prototypes and products. Areas in which our technology transfer eeort is mostly concentrated comprise: anti-intrusion systems for domestic surveillance; monitoring of critical areas in railway environments (crossings; unattended stations); monitoring of dangerous areas in manufacturing lines; environment control (landslides). While not all equally mature, these activities are good indicators of the level of popularity that automated visual monitoring has achieved near the general public. As the demands grow deeper and wider, our project supports the idea that quite a variety of relevant problems can already be attacked and solved with today's knowledge; others await for further development of the visual technology, and will come within reach as novel, more sophisticated methods are developed and tested. 10 An object detection system for outdoor environments employing the background updating algorithm in order to deal with illumination changes (e.g., evolution of daylight, day/night transitions, shadows) is shown in Fig. 5. (a) (b) (c) Figure 5: (a) Gray-level input image as taken from a standard surveillance camera positioned in an outdoor environment (the access to the IRST parking); (b) the background image currently estimated; (c) the absolute diierence image between the background and the input image. A system able to localize and track a person moving inside a room, focusing the attention on the target by pointing a (third) mobile camera (Fig. 6). (a) (b) (c) Figure 6: (a-b) Gray-level input images as taken from a pair of cameras set in our lab. In each image, the white box indicates the area corresponding to the person (detected by comparing the input image with the reference background image); (c) map of the room controlled by the two cameras; the dotted line represents the trajectory of the person in the room. 9 A surveillance system was developed within the Esprit Project P7809 (DIMUS), which is able to perform real-time classiication of the crowding level on the platforms of underground stations. The system employs a change detection scheme based on frame by frame diierences plus background subtraction (Fig. 3); a neural-fuzzy classiier provides the nal estimate of the crowding level 7]. (a) (b) (c) Figure 3: (a) Gray-level input image as taken from the camera in an underground station; (b) areas in which signiicant motion is …