Body Pose Estimation Integrated With Notational Analysis: A New Approach to Analyze Penalty Kicks Strategy in Elite Football

Abstract

Body orientation of football players has proven to be an informative resource related to successful penalty kicks. OpenPose is one of the most popular open-source pose estimation technologies. This study aims: (i) to verify whether OpenPose can detect relevant body orientation angles from video data of penalty kicks in elite football and (ii) to investigate the relationship between these body angles and observable behaviors analyzed via an observational system for penalty kick analysis in football (OSPAF) with the penalty taker and goalkeeper strategy. A total of 34 penalty videos, with standardized viewing angle, from the main European leagues (2017-2020) were analyzed. Relevant body orientation variables were selected for penalty kicks analysis and were extracted from video data through OpenPose technique. The OSPAF, previously validated by experts, was used. The mean confidence score of OpenPose measures was 0.80 ± 0.14. OpenPose Retest reliability values was 0.976 ± 0.03. Logistic regressions were performed to investigate the relationship between OpenPose investigated variables (penalty taker: shoulder, hips, and nonkicking foot orientation; goalkeeper: right and left foot, anticipation), observable behaviors (OSPAF variables), and the strategy (penalty taker: goalkeeper dependent or independent; goalkeeper: shooter dependent or independent) in penalty kicks. The selected body orientation angle (goalkeeper anticipation) measured through OpenPose correlated significantly with the goalkeeper strategy. The prediction model of the goalkeeper's strategy had its accuracy increased to 97% when the variable goalkeeper anticipation was included [ ${\chi }_{\left(35\right)}^2$ = 49.648, p < 0.001]. Lower degrees of goalkeeper anticipation, the goalkeeper tactical action (awaiting), and run up speed (slow) were associated with a kicker-dependent strategy. Regarding the penalty taker, the selected body angles measured through OpenPose did not associate significantly with the shooter strategy. Body orientation analysis by using OpenPose has shown sufficient reliability and provides practical applications for analyzing the strategies adopted by goalkeepers in penalty kicks in elite football.

Keywords: OSPAF; OpenPose; body orientation; human movement; motion capture; performance analysis; soccer analytics.

Figure 1

Penalty kick viewing angle, frames analyzed, and process of pose estimation. The upper image corresponds to the moment when the penalty taker starts the run-up approaching the ball, and the down one corresponds to the moment when he touches the ball. The coordinates belonging to shoulders, hips, and non-kicking foot of penalty taker are mapped in a 2D field. LR-side Booleans (LR_Sh, LR_Hi, LR_SF: penalty kicker's shoulder, hips, and non-kicking foot orientation, respectively), angles (α_Sh α_Hi, α_SF: penalty kicker's shoulder, hip, and non-kicking foot, respectively) and confidences (C_Sh, C_Hi, C_SF: confidence score of the orientation of penalty kicker's shoulder, hips and non-kicking foot, respectively). Coordinates belonging to the neck and hip of the goalkeeper at these two moments are mapped to a vector and the angle is calculated (α_GK). D_GKR and D_GKL, the movement distance of right and left foot; C_GK, represent confidence score.

Figure 2

Lince Plus interface. Example of criteria and categories coding.