Automated viewer-centric personalized sports broadcast

Abstract

Mobile handheld devices and online media are currently reaching a stage where it is possible to watch streaming video/TV on them, such as a live sports game broadcast. However, current sports broadcast methods are designed to replicate the same feed to many different TV viewers, which makes it difficult to personalize the feed for each individual viewer. Clearly, since TV viewers have different tastes, this generic broadcast is not likely to satisfy the preferences (e.g., favorite player, specific camera angle) of every viewer. Advertisers already target their audiences with commercials during sports broadcasts in a personalized way (e.g., targeting commercials to specific demographics that are the expected viewers of a game), and it seems reasonable to expect the next step to be the personalization of the sports broadcast feed itself in a viewer-centric way. Adding human directors to create different personalized feeds in the broadcast-production trucks will not scale due to the associated costs. We propose instead a solution that uses a computer program to target the sports broadcast to suit the preferences of each viewer. In this paper, we explore designing an automated computer-driven broadcast director that provides different, personalized automated broadcasts, depending on the viewer's preferences as well as the specific actions unfolding in the game. By using a combination of video analysis to detect the relative player locations and "size" of player templates to predict the next few seconds of game action, along with viewer-specified preferences (e.g., favorite player, favorite camera angle), we automatically select the most interesting camera angle to display to that viewer in real-time. We present the preliminary experimental results of our prototype system using the different video-camera feeds of both an ice-hockey and an American football broadcast.