DEVELOPMENT OF TECHNOLOGIES AND METHODS OF ANALYSIS FOR THE FUNCTIONAL EVALUATION OF PERFORMANCE IN FOOTBALL

Football performance analysis has traditionally relied on expensive multi-camera tracking systems and labor-intensive manual video analysis, creating significant barriers to accessing comprehensive performance data across different competitive levels. This thesis presents the development and validation of automated performance analysis algorithms integrated within the XSEED smart shinguard system, combining Global Navigation Satellite System (GNSS) positioning with high-frequency Inertial Measurement Unit (IMU) sensors to capture both athletic and technical performance metrics in ecological match conditions. The research addresses three fundamental challenges in wearable-based football analytics: achieving accurate physical performance monitoring comparable to established tracking systems across diverse environmental conditions by developing novel IMU-based algorithms that enable continuous athletic assessment even in GNSS-denied scenarios; developing reliable automated detection and classification of technical events to reduce dependency on manual video analysis; integrating spatial context with biomechanical data to provide tactically meaningful performance insights. The XSEED system employs a comprehensive sensor architecture featuring dual-core ARM microcontrollers, 9-DoF IMU sensors sampling at 200 Hz, and 10 Hz GNSS receivers, enabling continuous data collection during official matches and training sessions. Advanced machine learning (ML) algorithms, including CNN+BiLSTM neural networks, were trained on accelerometer and gyroscope data for automated technical event detection, achieving 92% macro F1-score for adult populations and 89% for youth. Geofencing algorithms integrate spatial probability models trained on over one million professional football events from OPTA Sports data, classifying the detected technical event in shots, crosses, passes and long balls with 83% F1-score in both 11-a-side and 7-a-side pitches through Bayesian fusion of IMU data and spatial features. The ball speed estimation through ML models achieved RMSE of 3.21 km/h in controlled laboratory conditions and 13.6 km/h in match scenarios. Athletic performance validation demonstrated step detection accuracy of 95.8% and distance estimation with RMSE of 17.6 meters over 202-meter tracks across soccer-specific movement patterns. Novel applications include automated gait asymmetry assessment for injury prevention and neuromuscular fatigue monitoring through jump analysis algorithms. The system's practical impact was validated through partnerships with Major League Soccer clubs (Montreal CF, Philadelphia Union, Columbus Crew) and integration with professional video analysis software (VideoMatch, SportsCode). Automated highlight generation and XML synchronization reduced manual analysis time by approximately 70% while maintaining professional-standard accuracy. The research establishes a unified framework for democratizing access to comprehensive football performance analysis, enabling technical and athletic assessment across all competitive levels while enhancing rather than replacing existing analytical workflows.

Grantor dc:publisher

Università degli Studi di Milano

Year dc:date

2026

Author dc:creator

• SANTICCHI, GABRIELE

Contributors dc:contributor

• tutor: F. Esposito
• M. Zago
• A. Comi
• COMI
• ALDO
• G. Santicchi
• ESPOSITO, FABIO

• Wearable sensor
• football performance analysi
• machine learning
• technical event detection
• sports analytic
• IMU
• GNSS
• automated coaching tools
• Settore MEDF-01/B - Metodi e didattiche delle attività sportive
• Settore IINF-05/A - Sistemi di elaborazione delle informazioni

Statement dc:rights

• info:eu-repo/semantics/embargoedAccess
• license:Creative commons
• license uri:http://creativecommons.org/licenses/by-sa/4.0/

Language dc:language

eng

Handle dc:identifier

https://hdl.handle.net/2434/1230215

OAI identifier oai:identifier

oai:air.unimi.it:2434/1230215