Automatic Virtual Camera View Generation for Lecture Videos
Derek Pang, Sameer Madan, Serene Kosaraju, Tarun Vir Singh · 2010
ClassX, an interactive online viewing system developed at Stanford University, currently offers automatic tracking of the lecturer to create a virtual camera view for students. However, a tracking-based camera view does not mimic a human operator naturally. To address this issue, we propose a new learning algorithm to automatically generate a professional virtual camera view by learning the behavior of a human camera operator. Our algorithm is based on a human saliency model which predicts the viewer’s center-of-attention and additional features which drive the cinematic decision-making process of the system. Experimental result reveals our system can provide an effective camera view that is almost indistinguishable from a humangenerated camera view. 1.