Human Action Classification Based on Pose Estimation and Artificial Neural Network

Human computer interaction is an interesting topic in the current research work. Human action is useful for interaction with the computer. However, human action classification is a challenging topic. In this paper, we present the human pose classification using artificial neural network. There are ten classified actions including Both Hands Up, Right Hand Up, Left Hand Up, Both Hands Bye, Right Hand Bye, Left Hand Bye, Both Hands Fly, Right Hand Fly, Left Hand Fly, and Both Hands Down. We apply pose estimation to identify the body’s key points with results in 2-dimensional data as the basis of human skeleton. The action model is generated from the human skeleton by employing an Artificial Neural Network. The data has been processed by classification model with an accuracy of 95.7 percent, a precision of 96.2 percent, a recall of 95.8 percent and F1-score of 96.5 percent.