姿勢推定技術を活用した人物の動作認識手法の精度評価

抄録

In this paper, we evaluate recognition accuracy of human motion by using a state-of-the-art pose estimation technology called “OpenPose”. We construct 3D character models as a training data by using 3D modeling software and train the 3D data by using Support Vector Model (SVM) or Neural Network (NN). Evaluation results indicate that NN provides a higher recognition accuracy compared to SVM.