Semi-automated Placement of Annotations in Videos
Vineet Thanedar, Tobias Höllerer · 2004
In this paper, we present a framework for the insertion and semiautomated placement of visual annotations into video footage. Visually overlaid annotations are very common in telecasts (e.g., statistics in sports broadcasts, banner advertisements), where they are strategically placed by hand, or incorporated into the video stream using elaborate camera tracking and scene modeling techniques. Such information is not commonly available for raw or unprepared videos. We look at the problem of automatically placing annotations within the space and time domains of unprepared videos using computer vision and image analysis techniques. We use measures of intensity and color uniformity, motion information, and a simple model of cluttered screen space to determine regions in the video that are of relatively minor importance to the human visual perception. We present a taxonomy of visual annotations, a subset of whose we realize as automatically placed video overlays, implemented in our prototype system named DAVid (Digital Annotations for Video). We conclude with an outlook of potential applications of these techniques in interactive television.