Abstract
Direct replay of the experience of a user in a virtual environment is difficult for others to watch due to unnatural camera motions. We present methods for replaying and summarizing these egocentric experiences that effectively communicate the user's observations while reducing unwanted camera movements. Our approach summarizes the viewpoint path as a concise sequence of viewpoints that cover the same parts of the scene. The core of our approach is a novel content-dependent metric that can be used to identify similarities between viewpoints. This enables viewpoints to be grouped by similar contextual view information and provides a means to generate novel viewpoints that can encapsulate a series of views. These resulting encapsulated viewpoints are used to synthesize new camera paths that convey the content of the original viewer's experience. Projecting the initial movement of the user back on the scene can be used to convey the details of their observations, and the extracted viewpoints can serve as bookmarks for control or analysis. Finally we present performance analysis along with two forms of validation to test whether the extracted viewpoints are representative of the viewer's original observations and to test for the overall effectiveness of the presented replay methods.