Improved YOLOv5 Algorithm for Special People Recognition in a New Guide Robot Dog for the Blind

Aiming at the problem of target recognition accuracy of the guide quadruped robot dog, this paper proposes a novel YOLO (SG-YOLOv5) detection and recognition algorithm based on changing the loss function and activation function. The proposed SG-YOLOv5 method replaces the activation function ReLU with SiLU and the loss function CIoU with GIoU in the YOLOv5 model. In addition, a database of special populations containing four categories, namely, blind family members, doctors, firefighters, and traffic police officers, is trained and applied to a real Unitree Go1 quadruped robot dog. The experimental results show that the recognition accuracy of the blind-guiding robot dog for special people reaches 94.3%, the check-all rate reaches 91.7%, and the average accuracy is as high as 95.1%, which verifies the effectiveness of the algorithm.