Open Access

Sinan Emre Süme, Katrin-Misel Ponomarjova, Thomas Wendt, Stefan Rupitsch

• This paper introduces the use of convolutional neural networks to detect and classify humans and robots in Human-Robot Collaboration workspaces based on their thermal radiation power. The measurement setup includes an infrared camera, two cobots and up to two persons walking or interacting with the cobots in industrial settings. The chosen architectures are the YOLOv5 and YOLOv8 in different modelThis paper introduces the use of convolutional neural networks to detect and classify humans and robots in Human-Robot Collaboration workspaces based on their thermal radiation power. The measurement setup includes an infrared camera, two cobots and up to two persons walking or interacting with the cobots in industrial settings. The chosen architectures are the YOLOv5 and YOLOv8 in different model sizes. The results are promising, showing real-time object detection in industrial settings with up to 303 fps with the YOLOv8n model. YOLOv5m achieves the best mAP50 result at 99.2% and the YOLOv5m achieves the best mAP50-95 at 85.8%…