The thesis addresses the development of a system for automatic error detection
in the execution of selected football exercises based on video recordings.
The system is based on human pose estimation and object detection,
which, combined with predefined rules for correct and incorrect execution,
enables the identification of errors in football exercises.
As part of the thesis, we analyzed the requirements of football coaches,
selected three representative exercises, and precisely defined the criteria for
correct and incorrect execution. We also examined existing related solutions
in the field of computer vision. The system’s performance was evaluated
in collaboration with football experts and coaches, who contributed to the
definition of correct and incorrect exercise execution and provided feedback
on the system’s operation.
The result of the thesis is a working system that successfully detects
common errors in three selected football exercises and enables visualization
of these errors for both coaches and players. In doing so, the system contributes
to more efficient and accessible training analysis without the need
for expensive equipment, and opens up possibilities for further development
and broader use in sports practice.
|
