We design and implement a networked visual monitoring system for surveillance. Instead of the usual periodical monitoring, the proposed system has an auto-tracking feature which captures the important characteristics of intruders. We integrate two schemes, namely, image segmentation and histogram comparison, to accomplish auto-tracking. The developed image segmentation scheme is able to separate moving objects from the background in real time. Next, the corresponding object centroid and boundary are computed. This information is used to guide the motion of tracking camera to track the intruders and then to take a series of shots, by following a predetermined pattern. We have also developed a multiple objects tracking scheme, based on object color histogram comparison, to overcome object occlusion and disocclusion issues. The designed system can track multiple intruders or follow any particular intruder automatically. To achieve efficient transmission and storage, the captured video is compressed in the H.263 format. Query based on time as well as events are provided. Users can access the system from web browsers to view the monitoring site or manipulate the tracking camera on the Internet. These features are of importance and value to surveillance.
Relation:
淡江理工學刊=Tamkang journal of science and enginnering 2(3), p.149-161