Modelling Tennis Rallies Using Information from both Video and Audio Signals.pdfVIP

Modelling Tennis Rallies Using Information from both Video and Audio Signals Gordon Hunter, Ahmed Shihab and Krzysztof Zienowicz Faculty of Computing, Information Systems and Mathematics, Kingston University, Penrhyn Road, Kingston-upon-Thames, KT1 2EE, U.K. G.Hunter@kingston.ac.uk , ais@kingston.ac.uk , k.zienowicz@kingston.ac.uk Abstract. Sports can contain events that occur too quickly to be captured by 25 frames-per-second video. An audio signal, sampled at a much higher temporal rate, provides a way to detect events over a shorter timescale. However, there are complications such as noise, echoes and latencies. Nevertheless, we have found statistical evidence that could allow an audio-based stroke detector to predict future events, allowing it to avoid potentially misleading sound event features. 1. Introduction The use of automated video systems for tracking balls or players in sports matches has been widely used for several years. For example, the Hawkeye system (Hawk-Eye Innovations, 2006) is used by TV companies during coverage of major tennis tournaments and international cricket matches to show animated ball trajectories and give viewers an objective view as to whether or not batsmen or a tennis shot had been correctly adjudged “Out” or “Not out” by umpires. The use of acoustic information, with the exception of Channel 4’s Snickometer (Hughes, 2005) – used in cricket to detect whether a ball has made contact with the bat, or part of the batsman’s clothing or protective equipment - has been far less common. In many bat-and-ball or racquet-and-ball sports, the ball is normally only in contact with the bat or racquet for around 5 milliseconds (Brody et al, 2002) – a timescale rather shorter than the duration of a single video frame – making the visual detection