Ball tracking for tennis video annotation

Abstract

Tennis game annotation using broadcast video is a task with a wide range of applications. In particular, ball trajectories carry rich semantic information for the annotation. However, tracking a ball in broadcast tennis video is extremely challenging. In this chapter, we explicitly address the challenges, and propose a layered data association algorithm for tracking multiple tennis balls fully automatically. The effectiveness of the proposed algorithm is demonstrated on two data sets with more than 100 sequences from real-world tennis videos, where other data association methods perform poorly or fail completely.