Soccer Last Touch and Automatic Event Detection with Skeletal Tracking Data

Abstract

With the rapid growth of soccer data collection technology worldwide, there has come about an increasing need for new efficient methods to analyze match data. This would help soccer stakeholders more easily and efficiently scrutinize game events for strategy improvement and individual player evaluation. Currently, most existing event data is annotated manually by hand, which is an extremely time-consuming task. Recent works in automatic event generation leverage decision tree algorithms to partially identify game events from player center of mass and ball tracking data, but have shown to be limited in accuracy in practice. New computer vision models have enabled the extraction of player joint data from video broadcast, providing a newer, richer dataset for automatic event detection. The proposed thesis will seek to validate brand-new skeletal joint data, determine the last player to touch the ball at any timestamp during a match, and build a decision tree algorithm for classifying duel-like events and goalkeeping outcomes with the additional context of player joint location.