Automatic annotation of tennis videos (AAOTV)

Abstract

Vision is a crucial must have for both humans and computers. As a general idea, Vision deals with the process of object recognition, objects localization in a specific space, tracking of objects of interest and also the recognition of certain actions which these objects exhibit. Computer Vision varies in some aspects when compared to human vision. This is due to the fact that computer vision is active while human vision is passive. Human vision relies on external sources to be efficient such as external energy sources which include sunlight, light bulbs and also fires which provide light that reflects the objects to our eyes. On the other hand computer vision is active since they can carry their own energy sources such as Radars. The basic idea of this thesis is to process a tennis video, taken via a static camera and to perform detection and tracking of both the tennis players and the tennis ball and finally produce annotations of the tennis game. In general words our work should successfully act as a Commentatory of a normal tennis match. The main steps involved in this process include a tennis court line detection to determine the coordinates of the lines of the court. Another module is an adaptive background subtraction technique, which is used to separate the background from the foreground and therefore detect the objects of interest. An important factor that this subtraction technique should have, is to successfully adapt to the changes in the environment, this is because the tennis match is played outside and therefore changes in the lightening are an a priori assumption. After that objects are detected the next thing is to successfully track their motion within the scene. In the case of my thesis the most important object to track is the tennis ball. By tracking the tennis ball important annotations would be known such as when the player stokes the tennis ball, when the ball bounced on the court, if the ball bounced outside the court and others. The result produced from our work should be a video with the ball tracked together with a set of annotations at certain intervals of the video. Our program should finally collate the tracked video and annotations together for the viewer to be able to see the tracked tennis video and the annotations in one view.