Analysis of work behavior using GP-HSMM-based double articulation analyzer

Abstract

In order to improve the efficiency of manufacturing, analyzing the data collected from sensors is required. However, it is not easy to automate the analysis process because of the variety of human operations and order during work. For this reason, analyzing is mainly conducted manually, which is time-consuming and labor-intensive. In this paper, aiming to automate the analysis, we propose a method to estimate the skeleton of workers with high accuracy from data captured by multiple RGB-D cameras and to analyze the estimated skeleton series based on the Gaussian Process Hidden Semi Markov Model Double Articulation Analyzer (GP-HSMM-DAA). In the experiment, we show that the proposed method can accurately estimate the skeleton of workers and extract repetitive motions and characteristic behaviors composed of repetitive motions.