Automated Classification of Changes of Direction in Soccer Using Inertial Measurement Units

Abstract

Changes of direction (COD) are an important aspect of soccer match play. Understanding the physiological and biomechanical demands on players in games allows sports scientists to effectively train and rehabilitate soccer players. COD are conventionally recorded using manually annotated time-motion video analysis which is highly time consuming, so more time-efficient approaches are required. The aim was to develop an automated classification model based on multi-sensor player tracking device data to detect COD > 45°. Video analysis data and individual multi-sensor player tracking data (GPS, accelerometer, gyroscopic) for 23 academy-level soccer players were used. A novel ‘GPS-COD Angle’ variable was developed and used in model training; along with 24 GPS-derived, gyroscope and accelerometer variables. Video annotation was the ground truth indicator of occurrence of COD > 45°. The random forest classifier using the full set of features demonstrated the highest accuracy (AUROC = 0.957, 95% CI = 0.956–0.958, Sensitivity = 0.941, Specificity = 0.772. To balance sensitivity and specificity, model parameters were optimised resulting in a value of 0.889 for both metrics. Similarly high levels of accuracy were observed for random forest models trained using a reduced set of features, accelerometer-derived variables only, and gyroscope-derived variables only. These results point to the potential effectiveness of the novel methodology implemented in automatically identifying COD in soccer players.

Ubiquitous Computing in Sports and Physical Activity—Recent Trends and Developments

“…The classification of human activities in sports performance can be improved when the sensors’ signals are used as input for machine learning algorithms. Recently, the ability of algorithms in distinguishing such datasets based on different conditions have been tested on canoeist’s level of expertise [ 5 ], different actions in table tennis [ 10 ], types of kick in Australian football [ 33 ] and change of directions in association football [ 11 ]. Together with the use of multiple fused sensors, a recent work has also tested different machine learning algorithms to recognise movement patterns.…”

“…The results of overall accuracy for 2-Kick and 4-Kick classification were higher (over 80%) when the random forest model was applied, while kick distances were more difficult to estimate with the classification models used (overall accuracy of 63%). Reilly and colleagues [ 11 ] developed an automated classification model to assess movements with changes of direction during competitive matches in association football. Players wore a multi-sensor device containing an augmented navigation satellite system and IMUs to provide position and attitude, and the random forest classification model was applied.…”

“…For methods of processing, current examples are described in Liu et al [ 8 ] and Hannan et al [ 9 ]. Finally, for analysis methods, examples are described in LeBlanc et al [ 3 ], Yanan et al [ 10 ], and Reilly et al [ 11 ].…”

The Design of GNSS/IMU Loosely-Coupled Integration Filter for Wearable EPTS of Football Players

“…Motion analysis becomes important for improving athlete performance and reducing athletes’ injury risk. IMU (Inertial Measurement Unit) sensors which consist of three-axis accelerometers, three-axis gyroscopes, and three-axis magnetometers have been used to estimate and provide the attitude, position, and velocity of athletes [ 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 ]. The head or foot injuries of sports players can be monitored by analyzing G-impacts and reaction forces using the measured acceleration data from IMU sensors [ 13 , 14 , 15 ].…”

“…The different IMU sensor positions can be possible to provide various physical load estimates of athletes and analysis the motion of athletes, i.e., football players movement intensity information [ 16 ], runners’ stride length and stride velocity, analysis at ground contacts [ 17 ], postural demands of professional soccer players [ 18 ], velocity measurements for team sports [ 19 ], and the analysis of foot swing at football kicks [ 20 ]. Deep learning techniques using IMU sensor information were also used to classify football activities [ 21 , 22 , 23 ].…”