Biomechanical analysis and model development applied to table tennis forehand strokes

Table tennis playing involves complex spatial movement of the racket and human body. It takes much effort for the novice players to better mimic expert players. The evaluation of motion patterns during table tennis training, which is usually achieved by coaches, is important for novice trainees to improve faster. However, traditional coaching relies heavily on coaches qualitative observation and subjective evaluation. While past literature shows considerable potential in applying biomechanical analysis and classification for motion pattern assessment to improve novice table tennis players, little published work was found on table tennis biomechanics. To attempt to overcome the problems and fill the gaps, this research aims to quantify the movement of table tennis strokes, to identify the motion pattern differences between experts and novices, and to develop a model for automatic evaluation of the motion quality for an individual. Firstly, a novel method for comprehensive quantification and measurement of the kinematic motion of racket and human body is proposed. In addition, a novel method based on racket centre velocity profile is proposed to segment and normalize the motion data. Secondly, a controlled experiment was conducted to collect motion data of expert and novice players during forehand strokes. Statistical analysis was performed to determine the motion differences between the expert and the novice groups. The experts exhibited significantly different motion patterns with faster racket centre velocity and smaller racket plane angle, different standing posture and joint angular velocity, etc. Lastly, a support vector machine (SVM) classification technique was employed to build a model for motion pattern evaluation. The model development was based on experimental data with different feature selection methods and SVM kernels to achieve the best performance (F1 score) through cross-validated and Nelder-Mead method. Results showed that the SVM classification model exhibited good performance with an average model performance above 90% in distinguishing the stroke motion between expert and novice players. This research helps to better understand the biomechanical mechanisms of table tennis strokes, which will ultimately aid the improvement of novice players. The phase segmentation and normalization methods for table tennis strokes are novel, unambiguous and straightforward to apply. The quantitative comparison identified the comprehensive differences in motion between experts and novice players for racket and human body in continuous phase time, which is a novel contribution. The proposed classification model shows potential in the application of SVM to table tennis biomechanics and can be exploited for automatic coaching.