Semi-automated Placement of Digital Annotations in Video Feeds


Semi-automated Placement of Digital Annotations in Video Feeds

Vineet Thanedar          Tobias Höllerer

Screenshots & Videos Publications


The augmentation of real video feeds with virtual elements is an important aspect of applications in areas such as real-time broadcasting and augmented reality. The real-time overlay of game scores, sponsor logos, product advertisements, 2D labels in AR applications are some examples. In most applications, the placement of annotations is governed by a priori knowledge about the scene (virtual models), camera/object tracking information, and/or a human operator. Consequently, one or more the above mentioned elements of information is required for the optimal placement of annotations in video feeds.

In this project, we address the automated placement of annotations in video feeds where no such information is available. We approach the problem from an image-based perspective to compute relatively unimportant or less interesting regions in the video for annotation placement. We establish a few elementary properties that are distinctly perceivable to the human visual system. We identify uniformity, motion, and clutter as Elementary Perceptual Descriptors and apply quantitative measures for each to analyze different regions in a video. We are presently developing a framework and a complete prototype system that enables the real-time visual placement of annotations in arbitrary video feeds.