An automatic visual analysis system for tennis. (December 2013)

• Record Type: Journal Article
• Authors: Connaghan, Damien; Moran, Kieran; O'Connor, Noel E
• Is Part Of: Proceedings of the Institution of Mechanical Engineers. Volume 227:Number 4(2013)
• Page Start: 273
• Page End: 288
• Publication Date: 2013-12
• Subjects: Sports coaching software -- instrumented coaching tools -- visual sensors -- automatic event detection; Sports sciences -- Periodicals; Sporting goods -- Design and construction -- Periodicals; Sports -- Technological innovations -- Periodicals; Biomechanics -- Periodicals; 613.71
• DOI: 10.1177/1754337112469330
• Languages: English
• ISSNs: 1754-3371

Abstract

This article presents a novel video analysis system for coaching tennis players of all levels, which uses computer vision algorithms to automatically edit and index tennis videos into meaningful annotations. Existing tennis coaching software lacks the ability to automatically index a tennis match into key events, and therefore, a coach who uses existing software is burdened with time-consuming manual video editing. This work aims to explore the effectiveness of a system to automatically detect tennis events. A secondary aim of this work is to explore the benefits coaches experience in using an event retrieval system to retrieve the automatically indexed events. It was found that automatic event detection can significantly improve the experience of using video feedback as part of an instructional coaching session. In addition to the automatic detection of key tennis events, player and ball movements are automatically tracked throughout an entire match and this wealth of data allows users to find interesting patterns in play. Player and ball movement information are integrated with the automatically detected tennis events, and coaches can query the data to retrieve relevant key points during a match or analyse player patterns that need attention. This coaching software system allows coaches to build advanced queries, which cannot be facilitated with existing video coaching solutions, without tedious manual indexing. This article proves that the event detection algorithms in this work can detect the main events in tennis with an average precision and recall of 0.84 and 0.86, respectively, and can typically eliminate manual indexing of key tennis events.