Event detection in spatiotemporal soccer data using neural networks

• Main Author: Wilbert
• Other Authors: Cheng Long
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; Engineering::Computer science and engineering::Computing methodologies::Pattern recognition
• Online Access: https://hdl.handle.net/10356/148013
• Institution: Nanyang Technological University

In the world of professional soccer, data analytics of individual player and team performance are important for the success of a team. These performance analyses are based on positional data and sequence of events occurring in the match (e.g. kick, pass, shot). The positional data of the ball and players are collected automatically using cameras and sensors. However, the events are still identified manually by soccer analysts, which can be time-consuming and error-prone. This project implements and evaluates three-layer and four-layer artificial neural networks (ANN(3) and ANN(4)) as well as recurrent neural networks (RNNs) for automatic event detection in spatiotemporal soccer data. The datasets used were derived from the synthetic SoccER (Soccer Event Recognition) dataset, which was originally generated using modified open-source soccer simulation game engine. As inputs for the neural networks, several time-dependent features were used and calculated from the positional data of the ball. The evaluation of the test results showed that the neural networks were able to detect events accurately, particularly for ‘BallPossession’ and ‘KickingTheBall’ events. Among the neural networks tested, ANN(4) achieved the best performance across all events. However, all neural networks were unable to detect ‘Tackle’ event successfully. Nonetheless, the findings from this project still further verify that using neural networks is a highly suitable approach to perform automatic event detection in spatiotemporal soccer data. This can improve the effectiveness and accuracy of event annotation process to support performance analyses in soccer.