Description

Predicting fouls in soccer remains challenging despite recent advances in computer vision techniques for player tracking and pose estimation. The small size of the player images and the need to analyze subtle interactions complicate the task. This study presents an approach that aggregates multiple views of actions, focusing on detected players. By integrating video data and bounding box positions from the SoccerNet-MVFoul dataset, our model utilizes a combination of multilayer perceptrons and decoder networks to enhance detection accuracy.