This project is part of an Off-Side detector project, and our mission was to find players and identify them by color criterion.
Abstract
This project is part of an Off-Side detector project, and our mission was to find players and identify them by color criterion.
The input we receive is an avi format movie of a soccer game generated by a static camera, the teams shirt’s color and some other technical parameters.
The output we produce is two avi format movies, for each team (color), where the players of the team are marked with a half transparent box.
Below are examples for an input frame from a soccer game movie (right), and a frame from the output movie we produced (left).
In this particular example, the yellow color was detected.
The algorithm
In order to solve the problem of soccer players recognition, we have created an algorithm, based on basic methods in image processing.
First, we analyzed a still frame, and understood the characteristics of the frame. We came to a conclusion that finding players of any color can be done by using the facts that players are mobile, and significant from their surrounding, and thus using diff picture of time domain, and edge picture of space domain.
The next step was to divide the players into two groups, based on their color. This task was harder and we had to look for more complex solutions.
The color recognition was done by transforming a frame to HSV color space (will be detailed below).
Finally, in order to improve the quality of the results, we made a cross check for every frame with the previous frame, knowing that the changes between two succeeding frames are not dramatic.
The algorithm is presented in the following block diagram and detailed on the next page:
HSV color space
The HSV color space is much more efficient for color recognition, because while in RGB color space, all three values are used to define a color, in HSV color space, only one value (H), is used.
The following is a typical scene from the soccer movie we used:
The following are the gray level histogram versus the V value histogram of the HSV space of this scene:
Results and Conclusions
The goal was partially achieved. The results show that we succeeded in recognition at rate of 75.4% for yellow players, and 87.7% for red players, not as good as needed to develop a commercial system, however, we think that for better results, one should take some steps, and take into account some facts:
1. Try to improve the quality of the input movie.
2. Develop a new, complex algorithm for color recognition.
3. Solve exception problems, such as referee recognition etc.
4. Our algorithm is strongly depended on the input movie.
5. The recognition has high standard deviation, and thus, it’s wise to come with a more stable solution.
Acknowledgments
We would like to thank our supervisor, Guy Gilboa, for his guidance and support of this project. We would also like to thank the staff of the VISL lab for their help.
Also we would like to thank The Ollendorff Minerva Center Fund which supported this project