Automatic Detection of Serve Actions in Tennis Videos

The analysis, annotation, and indexing of sports videos are still a very significant and interesting challenge in the computer vision area. It is due to the fact that sports videos are the most frequently watched videos in the Internet, widely broadcasted in special TV channels, and presented every day in sports news. Videos of the most popular sports disciplines such as football, soccer, basketball, volleyball, ice hockey, or tennis are particularly studied in theoretical investigations as well as tested in experimental research. The aim of the analysis of sports videos is, among others, temporal segmentation and extraction of highlights. In tennis, which is very popular and eagerly watched, the service is a shot that starts every point in the tennis game, therefore, serves are natural and obvious boundaries of events and can play the role of cuts in editing videos making the temporal segmentation of continuous tennis video possible. Then some segments can be chosen from the video segmented in this way as the highlights, for example tennis aces. The goal of this paper is to present a method of detecting serve actions as highlights in tennis videos. The proposed framework is based on a neural network classification preceded by the recognition of playing fields and players’ positions on a tennis court as well as the analysis of players’ silhouettes. The tests performed have confirmed the efficiency of the proposed framework of the automatic detection of serves in tennis matches.