Robust tracking for interactive social video

We propose a novel approach for tracking video objects in interactive multimedia systems. Instead of designing a single tracking algorithm that works well with some templates (video objects to be tracked) but fails with others, our adaptive algorithm uses three distinct tracking techniques with different strengths and weaknesses. The tracking technique that is selected for a given template depends on its visual content and the content of the video. Our approach is robust and allows tracking requests of users to be computed despite an imprecise selection of a template. In addition, our approach is designed to comply with high precision, speed, and scalability to support many users simultaneously. As an exemplary testing environment for this tracking system, an interactive hypervideo system was chosen. The system was integrated into the social network Facebook and used by more than 12,000 users.