By: Steve Immanuel Harnadi, Zachrandika Alif Syahreza, M. Ibnu Syah Hafizh, Parnaek R. Siagian, Gian Denggan Bendjamin, Thomas Stefen Mardianto, Suhono Harso Supangkat
Indonesia, as the one of biggest country in South East Asian region, has lots of potential players playing in football’s national team. Nevertheless, the football’s national team achievement still needs lots of improvement because current Indonesia’s FIFA world ranking is only in 149th from total 200 countries joining FIFA federation. The ranking’s position is overally still under another teams from South East Asian region such as Thailand, Vietnam, Philippines, or even Malaysia. Hence, according to the background, we develop the project to contribute for football’s national team development as Integrated System for analyzing match in Indonesia overally and also player’s performance.
Smart Football’s Integrated System consisted of another smaller components (projects) interrelated each other, with end of product provided in Website’s Dashboard. The small projects, except the dashboard, are mainly developed using machine learning algorithms to analyze the football’s match videos while another project using event tracking data retrieved from video analytics before to evaluate the player’s performance in team formation. Hereby are the short descriptions of project’s components covered in the Smart Football’s Integrated System:
Object & Event Detection Model
Object Detection model aims to recognize object in football for identifying, tracking, and classifying the object for each frame along the match videos in football using YOLOv8 library. The datasets for constructing this model are retrieved from SoccerNet dataset and DFL League datasets. The experiment’s result of this model is depicted in figure (a). Meanwhile, Event Detection model aims to detect the actions in football’s match videos using technique named Temporal Context Aggregation. Overall performance of the model and its actions detected is described in figure (b), while figure (c) describes the result of the project in the targeted video.
Expected Goals Model
Expected goal (xG) is one of the most common metrics used in analyzing football matches. It is usually used to measure the likelihood of a shot turning into a goal. By using historical shot data and machine learning concepts, the probability of an attempt becoming a goal can be calculated. To implement xG calculation of the player/teams, this module use best gradient boosting algorithms to output the xG value. The figure (d) is the example of xG representation of the team visualized in the dashboard.
Localization Model provides an in-depth analysis of football’s status in Indonesia, its growth potential, the significance of strategy and teamwork, the role of the coach, and the limitations of current game analysis methods. It also explores the potential of data analysis and artificial intelligence (AI) in enhancing game analysis, with a particular focus on player localization technology, which utilizes camera calibration and field detection concepts (see figure (e)).
Player’s Performance Model
Player’s Performance model use the event tracking data acquired from Object Detection Model, Event Detection Model, Expected Goals Model, and Localization Model (consideration in future features). For further steps, the event tracking data is analyzed using team formation and stochastic process algorithms to evaluate which player suitable to fill the position in football’s team formation. See figure (f) for example of event analysis conducted as preliminary step for constructing the model.
Website’s Dashboard Statistic
Smart Football Website is a web platform that provides an in-depth analysis of the world of football. With detailed and comprehensive match data visualization, this website provides statistical information, expected goals (xG), heatmaps, and key events in the match. By consuming data from above machine learning’s models, this website is also useful in providing information to football fans about the performance of teams during matches (see picture (g) for interface example).