This study aims to quantitatively identify goalkeeper defensive strategies and performance determinants during soccer penalty kicks by integrating AI-based 2D pose estimation with kinematic analysis using real broadcast match footage. The research was...
This study aims to quantitatively identify goalkeeper defensive strategies and performance determinants during soccer penalty kicks by integrating AI-based 2D pose estimation with kinematic analysis using real broadcast match footage. The research was conducted in two stages. In Study 1, joint coordinates obtained through MMPose-based pose estimation (utilizing the pose_hrnet_w48_dark+ model) from single-camera broadcast videos were compared with expert manual digitization to examine the feasibility of an AI-based markerless analysis pipeline. Results indicated an extraction rate of 83.48% and a Root Mean Square Error (RMSE) of 0.1847m. These findings demonstrate that the AI-based system achieves a level of accuracy suitable for high-level match analysis.
In Study 2, the validated pipeline was applied to actual match data (60 trials) to classify goalkeeper strategies into 'Anticipation' and 'Delayed'. Differences in temporal variables (Reaction Time [RT], Propulsion Time [PT]), displacement variables (Horizontal Displacement [HDT], Vertical Displacement [VDT]), performance variables (Peak Velocity [PV], Frontal Step Distance [FSD]), and a Save Difficulty Score (SDS) were analyzed. To account for the repeated-measures structure, Linear Mixed Models (LMM) were employed for statistical verification999999999.The results indicated that strategy differences were most clearly characterized by RT. The delayed strategy initiated movement an average of 118.22ms later than the anticipation strategy (p<.001). However, save outcomes were not significantly explained by the mere quickness of RT (p=.623). Instead, success was more directly associated with the final space occupancy (HDT, VDT) and whether sufficient time was available to generate propulsion (PT). Specifically, successful saves exhibited significantly longer PT (p=.033), greater HDT (p=.032), and higher VDT (p=.018) compared to failed attempts. Furthermore, the SDS index-composed of ball speed and distance-moderated the relationship between strategy and outcome (p=.012). This suggests that while a delayed strategy is effective under certain conditions, it becomes rapidly disadvantaged as the task difficulty increases.
In conclusion, penalty kick defense should be understood as a multidimensional task integrating cognitive judgment (information pickup and decision-making), tactical interaction (response to the kicker), and biomechanical execution (propulsion and space occupancy). By distinguishing strategies through RT and demonstrating that the key mechanism for success lies in space coverage and propulsion time, this study provides empirical evidence to inform applied coaching practices and future biomechanical research designs.