Standing Handball Throwing Velocity Estimation with a Single Wrist-Mounted Inertial Sensor

Creative Commons License


ANNALS OF APPLIED SPORT SCIENCE, vol.8, 2020 (ESCI) identifier identifier

  • Publication Type: Article / Article
  • Volume: 8
  • Publication Date: 2020
  • Doi Number: 10.29252/aassjournal.893
  • Journal Indexes: Emerging Sources Citation Index (ESCI), Scopus
  • Keywords: Inertial Sensor, Tri-axial Accelerometer, Machine Learning, Throwing Velocity, TEAM-HANDBALL, AUTOMATIC DETECTION, MEASUREMENT UNIT, BALL VELOCITY, KINEMATICS, ACCURACY, BASEBALL, EVENTS, ARM
  • Dokuz Eylül University Affiliated: Yes


Background. It is well known that overarm throwing is one of the most performed activities in the handball. Shoulder and glenohumeral injuries incidence are high in handball because of both pass, and shooting activity was executed repeatedly in high angular speed. Objectives. This study set out to investigate the usefulness of inexpensive commercial inertial movement sensors for prediction of throwing velocity in handball. Methods. After the IMU sensor (500 Hz) placed to the wrist of the dominant arm, players (n=4; 24.4 +/- 1.4 years, 181.75 +/- 11 cm height, 84.58 +/- 16 kg weight) performed 30 standing overarm throwing from a seven-meter distance with 1-minute rest between trials. Throwing velocity compared between radar speed gun and estimations of accelerometer data. Recorded acceleration data filtered (Butterworth 20 Hz 2nd order) than the acceleration vector magnitude calculated. Each throwing data aligned such as 125 data points of before and after the peak acceleration (250ms). Performance metrics of prediction models (Generalized Linear Model, Gradient Boosted Trees, and Support Vector Machine) calculated with root mean square, absolute error, and correlation coefficient parameters. Results. There were reasonably small absolute errors and root mean square values of the machine learning models. Also, there was a very high correlation between measured and predicted velocities with all three models. Conclusion. This is the first study to examined machine learning models to predict handball throwing velocity using a high-frequency triaxial accelerometer. The finding of the present study revealed that the wrist-attached accelerometer precisely estimates the throwing velocity in handball. Further research is required to quantifying the overarm activities in handball, which included block, defensive contact, passing, or shooting. Therefore, the accelerometer-based collected data may provide detection of movement in game-play automatically so that the upper extremity load of players can be monitored and avoid the possible overuse injury risk.