Real-time Generation of Annotated Video for Surveillance

We present a system that generates annotated video in realtime. Moving objects are first automatically extracted by means of video analysis and then coded using MPEG–4. Moreover, an MPEG–7 description of object features is generated. This description is used for automated event detection as well as for rendering. Rendering enhances important objects in the scene on the receiver side, thus facilitating the task of a surveillance operator. The performance of the system is demonstrated using four surveillance videos. Experiments show that objects enhancement can be achieved at low additional cost and illustrate how automated event detection is obtained by taking advantage of the physical scene description based on MPEG–7.