Real-time Person Detection and Tracking in Panoramic Video

The format agnostic production paradigm has been proposed to offer more engaging live broadcasts to the audience while ensuring the cost-efficiency of the production. An ultra-HD resolution panorama is captured, and streams for different devices and user profiles are semiautomatically generated. Information about person positions and trajectories in the video are important cues for making editing decisions for sports content. In this paper we describe a real-time person detection and tracking system for panoramic video. The approach extends our earlier tracking by detection algorithm by addressing a number of robustness issues that are especially relevant in sports content. The design of the approach is strongly driven by the requirement
 to process high-resolution video in real-time. We show that we can achieve improvements of the robustness of the algorithm while being able to perform real-time processing.