Metadata extraction and organization for intelligent video surveillance system

The research for metadata extraction originates from the intelligent video surveillance system, which is widely used in outdoor and indoor environment for the aims of traffic monitor, security guard, and intelligent robot. Various features are extracted from the surveillance image sequences such as target detection, target tracking, object's shape and activities. However, the trend of more and more features being used and shared in video surveillance system calls for more attention to bridge the gap between specific analysis algorithms and end-user's expectation. This paper proposes a three-layer object oriented model to extract the surveillance metadata including shape, motion speed, and trajectory of the object emerging in image sequence. Meanwhile, the high-level semantic metadata including entry/exit point, object duration time is organized and stored which are provided for the further end-user queries. The paper also presents the experiment results in different indoor and outdoor surveillance scenarios. At last, a comparative analysis with another traditional method is presented.