Motion Monitoring for Limb Exercise
Ziqi Li, Xiao Ma, Meizhen Liu · 2019
In this paper, we proposed a motion monitoring strategy for limb exercise. The proposed strategy combines three important elements of limb motion: the motion pattern, the number of repetitions, and the period of each repetition. Two methods are adopted to recognize the motion pattern: support vector machine and MoveNet, which is a deep neural network we proposed base on CNN and LSTM. A method combining zero-crossing detection and wavelet transform is used to count the number of repetitions and analyze the period of each repetition. The experimental results illustrate that the precision of workout-action identification is up to 97.71%, and the average error between the period calculated and the actual value is 4.03%.