Study of features for the automatic selection of cameras

Abstract

The main purpose of this master thesis is to have a first approach with techniques that could possibly be used to make the work of the director easier during a soccer game. To do so, a study of some features has been made.

This thesis presents an algorithm that has been thought to behaves like the director, with some simplifications. This algorithm can be separated in different modules that are detailed separately.

The overall algorithm is first introduced. The modules are introduced following some observations that have been made and justify those choices.

The features that have been tested, as well as the way to extract them, are detailed afterwards. The motivations that lead to choose those features among all the possibilities are also given. The quality of the extraction has also been tested to ensure that correct features would be used for the rest of the program.

The system to compute the score of each camera and to rank them according to this score is then presented. Once again the choice of the algorithm is explained and its implementation is discussed.

The final part consists in testing this program with different sets of features to try to make a difference between the ones that can be considered as useful and the others. The results are mitigated. Some features are encouraging but others seem to be completely useless.

A critical analysis of the results has been made and multiple possible improvements are proposed for those who are interested in improving the work that has been done here.