Fast dynamic video content exploration

Although the amount of surveillance video collected is increasing rapidly, the development of algorithmic tools to help an operator efficiently review this data remains a challenging problem. Video synopsis techniques summarize the active content within long surveillance videos in the form of short video clips in order to make review by an operator much more efficient. In this paper we present a novel form of synopsis browsing, based on a cyclical temporal pixel remapping scheme, with several key advantages over traditional methods. Most importantly, our method is not track-based and therefore works well in scenes for which it is difficult to maintain robust tracks of all components. The simple computational nature of our algorithm also allows the formation of synopsis frames on-the-fly, enabling a powerful interactive forensic browsing experience that we refer to as dynamic video content exploration. We also present a novel set of experiments using human subjects to quantify the efficiency of using synopsis tools for forensic video review tasks, finding roughly an order of magnitude decrease in operator review time compared to video fast forwarding.