Ping Pong Motion Recognition based on Smart Watch

Abstract

Smart watches have become one of the most representative devices in wearable devices because of their unique advantages such as integration, portability, reliability, stability, universality and low environmental dependence. At present, it is mainly used for the monitoring of health indicators such as human heart rate. Whole-body inertial sensing devices cannot meet the actual needs of the general public for virtual sports because of high prices and inconvenient wear. In this paper, a single piece smart watch is used to study the recognition of the most common actions in table tennis which is a kind of fast-moving sport and has many fans through an improved convolution neural network model. The final experimental results show that the recognition accuracy reaches 95.46%, which can basically meet the needs of amateurs' motionSports.

Creating a Scoring System with an Armband Wearable Device for Table Tennis Forehand Loop Training: Combined Use of the Principal Component Analysis and Artificial Neural Network

“…Studies have used various sensors and data analysis methods for table tennis observations, such as video analysis to track the ball’s trajectory during service [ 11 ]; force sensors and electronic circuits to monitor the net [ 12 , 13 ] and detect ball–net impact during service [ 13 ]; inertial sensors mounted on the paddle to assess the ball’s speed and spin [ 14 ], estimate the trajectory of the paddle [ 15 ], and detect the type of stroke [ 16 , 17 ]; and inertial sensors worn on the limbs to detect shots [ 18 ], estimate kinematic parameters [ 19 ], and recognize stroke motions [ 20 , 21 , 22 , 23 , 24 , 25 , 26 ]. Machine learning classification algorithms are a reliable approach for recognizing basic human motions.…”

“…Machine learning classification algorithms are a reliable approach for recognizing basic human motions. The authors of related studies have recommended various expert machine learning models for table tennis observations, such as the k-nearest neighbor algorithm [ 20 , 21 ], support-vector machines (SVMs) [ 13 , 14 , 16 , 17 , 21 , 24 ], neural networks [ 17 , 18 , 22 , 23 , 25 ], or Long Short-Term Memory deep learning methods [ 26 ], all of which can achieve a sufficient level of accuracy for recognizing and classifying table tennis strokes. Because of their excellent nonlinear mapping and learning capabilities, neural networks can fully link information into network nodes and create a network mod-el with which to produce an effective prediction model.…”

Racquet Sports Recognition Using a Hybrid Clustering Model Learned from Integrated Wearable Sensor

“…Wang et al [26] used a two-layer Hidden Markov Model (HMM) to identify 14 types of badminton movements. Fu et al [27] used convolutional neural network models to identify and analyze the ping-pong movements based on inertial sensing data. However, the above methods cannot provide timely feedback information to users.…”