In this modern day and age, many of us use the television and radio to enable us to experience sporting events which we cannot physically attend. It is a common practice to employ a professional sportscaster to describe the plays which have been made, comment on strategy, and discuss the game. Unfortunately, when humans learn this skill, they also acquire an overwhelming urge to spout phrases such as, "Oh, the emotion...", which nauseate the hell out of the rest of us. Perhaps if we could create a computerized sportscaster, there would be no need to endure the musings of human sportscasters.
This project is intended as a preliminary step in the creation of this computerized sportscaster.
This project will consist of two distinct phases, both of which will take place at a tennis game. The first phase involves tracking the tennis ball as it is hit back and forth by the players. I would like to be able to distinguish the ball from the background, and draw a box around the ball.
In the second phase of the project, I will play with the program I developed in the first part, and see what else I can do. I would like to try to remove the ball from the picture entirely, creating a scene consisting of two players swinging their rackets at nothing. If this is successful, I will attempt to track the players themselves. Can I tell when the players are swinging their rackets? Can I track them to figure out what kind of shot they are making?
For the first part of the project, identifying the ball, I will use image differencing to find the ball in the picture. One possible hang-up that may happen is that it may be difficult to get tennis footage with a "clean" background. Perhaps using optical flow, I might be able to differentiate between the fast-moving ball and a slow-moving background.
In order to remove the ball from the picture, I will most likely cut out the part of each frame which has the ball in it, and replace it with the corresponding area from the previous frame. I'm sort of stumped as to how I will go about tracking the players, or figuring out the shots that they are making. The experience I gain in doing the first two parts of the project will help with this aspect.
Last Modified: Apr. 29, 1998
Joe Bayes, jbayes@cc.gatech.edu