Towards Safe Human-Robot Interaction: A Pilot Study on a Deep Learning-Assisted Workspace Monitoring System

This pilot study aims to explore the potential of a deep learning-assisted workspace monitoring system in ensuring safety in both social and industrial human-robot interaction settings. For this purpose, two vision sensors are used to collect multi-view datasets from different perspectives, with a single participant involved in 12 defined movement scenarios. The Residual Network (ResNet 18), a deep learning model, is employed to detect upper body movements based on the collected datasets. The experimental results demonstrate the accurate prediction of upper body movements by the proposed approach. Furthermore, the results also indicate the potential integration of this approach, which utilizes multiple inputs from various sensors, with the existing system introduced in previous work to facilitate a more dynamic workspace monitoring system for safety purposes.