A Classification Method of Human Motions Based on ApEn and RF

Jiayu Chen, Xinv Zhu, Dong Zhou, Chuan Lv


According to the non-stationary, nonlinear and high dimensional characteristics of human motion caption data, a classification method for human motions based on approximate entropy (ApEn) and random forest (RF) is proposed in this paper. First, by preprocessing the motion capture data, key joints are selected to represent the motion instead of the all original joints. Second, the feature vector of each motion is extracted by calculating its ApEn. Third, the feature vectors are severed as the input vectors of the RF classifier for human motions recognition. Finally an experiment is conducted and the classification results achieve a high classification accuracy for 3 motion patterns (walk, run and jump), which confirms the highly accurate and effective performance of the proposed method.


Full Text:



  • There are currently no refbacks.