Fall Detection in SoccerNet Data

SoccerNet dataset is a benchmark for action spotting in soccer videos. The action that we are focusing on in this paper is the fall of players on the soccer field during a match. Falls are the leading cause of injury; however, research on fall detection tends to focus on the elderly and in medical environments. Falls are not a risk that only affects the elderly. In a study of young adults, 48% reported falling at least once in the previous 16 weeks. Of those falls, over a third of them were a result of sporting activity. The SoccerNet dataset fall detection task is to predict whether or not a person has fallen in a given video. The data consists of video footage of soccer matches, and the task is to classify each frame as either containing a fall or not. The SoccerNet dataset is a large-scale dataset for fall detection in soccer. It consists of over 1,000 hours of video footage of soccer matches. The dataset is challenging due to the variability in the appearance of falls and the fact that falls are often not in the camera's focus. The dataset we are using is the SoccerNet tracking challenge dataset. This dataset contains 110 different 30-second clips of soccer games recorded from a single camera angle. We used pose estimation and player tracking to create a deep-learning model for our detection.