Feature Extraction System during Bolt Tightening Work by Regression CNN and Classification CNN

This paper proposes a general-purpose system to extract motion features of workers during an operation at manufacturing sites from video footage. Currently, there are several systems that are used to extract motion for specific types of work. However, most of these systems are difficult to apply across many types of work. In order to make our system applicable to a variety of work at manufacturing sites, our system uses end-to-end deep learning models to extract motion features. We have applied our system to bolt tightening work using a spanner and have confirmed the effectiveness of our method.