Open Access

Thomas Wendt, Sinan Emre Süme, Amal Kaithavalappil Ajay

• Human-robot collaboration is a key factor in the future development of flexible and transforming industries, where seamless interaction between humans and robots is essential. To facilitate this, robots need to perceive their environment based on sensor data while maintaining accuracy and the anonymity of humans. In this paper, we analyze a 3D object detection algorithm-based approach forHuman-robot collaboration is a key factor in the future development of flexible and transforming industries, where seamless interaction between humans and robots is essential. To facilitate this, robots need to perceive their environment based on sensor data while maintaining accuracy and the anonymity of humans. In this paper, we analyze a 3D object detection algorithm-based approach for deployment in challenging industrial settings. We make use of a measurement setup with objects to grasp, a collaborative robot, and two sensor-fused Time-of-Flight sensors to analyze the performance of the VoteNet algorithm for 3D bounding box detection of humans. We gather domain-specific data to enable the algorithm to be adapted to industrial settings. The fine-tuned algorithm achieved an accuracy of 80.55% for the prediction of the human class with an inference speed of speed of 86.5ms.…