Soccer Analysis addressed a critical and unresolved challenge in soccer performance analysis: the disconnect between physical data (from GPS/accelerometer sensors) and tactical data (from manually coded video analysis). While physical tracking systems offer metrics such as sprint counts or total distance, they miss contextual tactical insights (e.g. passes, shots). Conversely, tactical data captured through video coding is labour-intensive and prone to inconsistency.The emergence of computer vision techniques (YOLO, OpenPose) suggests a path toward automation of tactical analysis – but practical adoption has been hampered by occlusion effects and unreliable player identification.
The key objectives of Soccer Analysis’s are:
1. Creating a hybrid AI-based solution that fuses computer vision and MEMS accelerometer data, combining the real-time positional context of sensors with the high-level visual semantics of video analysis.
2.Allowing for accurate, real-time identification of both player movements and tactical actions, enabling a seamless, integrated performance analytics tool.
Expected Impact
1. Player Identification:Develop real-time player identification algorithms using computer vision fused with MEMS data to overcome occlusion and ambiguity in video footage.
2. Player Actions Model: Built machine learning models (e.g., SVMs) to classify key player actions (kicks, headers) by combining OpenPose visual features with inertial sensor data.
3.Demonstrator Technology:Soccer Analysis is validated using real match and training data, with structured feedback from professional clubs in the UK and Europe.
