Motion Recognition of Workers Using Human Pose Estimation and Neural Network

Abstract

We introduce a method that classifies worker’s action using pose detection and neural network model. Classification model in this method can be trained with small amounts of data by using a trained pose detection model, which extracts pose data which is low-dimensional, from video data which is high-dimensional. We evaluate the performance of this method by measuring the classification accuracy for worker’s five actions in a simulated work environment.