A Ping-Pong Arm Action Recognition Method by Equal-Ratio Coefficient Filter
Shaowei Wang, Ji Qi · 2017
A method is proposed to recognize realtime ping-pong player's action in Microsoft Kinect Sensor environment. A model containing three angles is established to represent ping-pong arm action features through skeletal data. Constructing an equalratio filter to calculate the likelihood probabilities for every action type in every skeletal frame data. The filter considers most poses and employs a coefficient to adjust the weights of the arm poses. The experimental results show that this method not only keep high real-time performance but also has high accuracy.