On-line video synchronization based on visual vocabularies

Abstract

This paper presents the procedure for on-line visual-content-based video synchronization. The motivation of our pioneering work is the existence of several off-line video processing systems employed in video classification or summarization applications, but no evidence of on-line solutions for video analysis. In some applications, the video streams go through the broadcast systems that delay the original video and also distort the original signal. The system that would be able to automatically detect such delays, transmission errors, distortions or broadcast failure is highly required. Our solution employs visual vocabularies that allow signing the video frames by bagof- words. The synchronization procedure is then based on searching for similar frames from different video streams. This paper also overviews the state-of-the-art techniques required for visual vocabulary building and discuss the convenient properties of techniques for real-time on-line systems.