BIOMECHANICAL ANALYSIS OF SPRINT RUN IN PROFESSIONAL SOCCER PLAYER USING ARTIFICIAL INTELLIGENCE PROCEDURES

Abstract

Sprint biomechanical analysis provides valuable information for performance assessment and injury prevention. However, most motion analysis systems require controlled laboratory environments and specialized staff, limiting their application in field sport settings such as soccer. This work presents VideoRun2D.v4, an automated markerless application capable of extracting biomechanical information ¿joint coordinates, angles, gait events, and biomechanical variables¿ from sagittal-plane sprint videos recorded with a conventional camera, through a user-friendly graphical interface. The system was implemented as a Streamlit web application built upon the MMPose framework, using the RTMPose model for pose estimation. It integrates the full pipeline into five sequential tabs, from video upload to automatic gait event detection and the extraction of 27 biomechanical variables per stride cycle. The system was evaluated across three sprints from ten soccer players, comparing the extracted variables against Kinovea through a mixed-design MANOVA in SPSS. Joint angle variables showed good agreement, with mean absolute errors between 4° and 8°, while temporal variables presented moderate errors (19 ms). Angular velocities remained the main limitation, showing high relative errors due to the noise amplification inherent in numerical differentiation. No significant effects of sprint repetition or leg were found for the majority of variables, confirming the consistency of the measurements. Overall, VideoRun2D.v4 demonstrates the feasibility of automated, markerless sprint analysis from a single camera under real-world conditions, offering an accessible alternative to laboratory-based systems.