Extraction of high-level video content for advanced video applications

Automated and effective techniques to extract video content such as moving objects and related semantic features have many applications in new video technologies. Video-based technologies include video database, video surveillance and video delivery over mobile terminals. Commentators predict that video surveillance (e.g., automated detection of unauthorised access) will grow dramatically in the coming years. The introduction of next-generation mobile services will make video contents accessible via mobile terminals. Automated content extraction would significantly facilitate the use and reduce the costs of these applications. This paper proposes a system for stable real-time extraction of high-level video content. The system consists of four interacting levels: enhancement to estimate and reduce noise, stabilisation to compensate for global changes such as global motion, analysis to extract moving objects, and interpretation to extract semantic features. Our system represents a video in terms of moving objects and related semantic features such as events. To achieve higher applicability, content is extracted independently of the context of the input video. Our system, implemented on over 6000 images with multiobject occlusion and artifacts, produces stable results in real-time. This is due to the adaptation to noise, the compensation of estimation errors at the various processing levels, and the division of the processing system into simple but effective tasks.