Efficiency analysis of deep learning-based object detection for safe human robot collaboration

Abstract

Nowadays, deep-learning object-based detection is very often used for predicting and ensuring safety, in workspaces shared by human operators and robots. The main challenge is to evaluate the accuracy of such models depending on the acquired data, selected Artificial Intelligence architecture and on the relevant parameters for achieving the best efficiency. In this paper, the efficiency of detecting deep learning-based objects, applying the Region-Based CNN (YOLOv8 Tiny) was analysed for detecting objects within human robot collaboration (HRC).The relevant parameters were selected in each area discussed and the efficiency of the YOLOv8 Tiny applied for the object’s recognition was analysed. The research results indicate that the techniques for detecting objects, in HRC, applying YOLOv8 Tiny for interferences in the recorded material analysed the most was achieved by some 90%, however, in the context of the analysis of smoke disruption this was insufficient.