Data-driven Summarization of Broadcasted Cycling Races by Automatic Team and Rider Recognition

Steven Verstockt, Alec Van den broeck, Brecht Van Vooren, Simon De Smul, Jelle De Bock

Abstract

The number of spectators for cycling races broadcasted on television is decreasing each year. More dynamic and personalized reporting formats are needed to keep the viewer interested. In this paper, we propose a methodology for data-driven summarization, which allows end-users to query for personalized stories of a race, tailored to their needs (such as the length of the clip and the riders and/or teams that they are interested in). The automatic summarization uses a combination of skeleton-based rider pose detection and pose-based recognition algorithms of the team jerseys and rider faces/numbers. Evaluation on both cyclocross and road cycling races show that there is certainly potential in this novel methodology.

Download


Paper Citation