Enhancing Underwater Human-Robot Communication: Real-Time Hand Gesture Recognition Using YOLO Models

To manage the challenge of diver safety, ROVs can be used as diver buddies. This requires communication between diver and the underwater robot. To aproach this problem this paper presents a novel underwater gesture recognition system for real-time hand gesture recognition based on YOLO models. The main contributions of this work are as follows: (1) the development and incorporation of a self-collected underwater gesture dataset, named Dive Autonomous Buddy (DAB), containing important gestures. (2) training of several YOLO models (YOLOv5x, YOLOv7, YOLOv8n, YOLOv9s, and YOLOv10s) for the purpose of gesture recognition; and (3) a comprehensive assessment and comparison of YOLO models in terms of underwater hand gesture recognition using metrics like precision, recall, F1-score and mean average precision (mAP). The results of this study show that YOLO-based systems can improve humanrobot communication underwater.