Restricted Area Sign Detector Using YOLO v5

Abstract

This paper presents the implementation of a YOLO v5 model in a mobile robot for the detection of restricted area signs and the delivery of medicines. The objective is to enhance privacy by preventing the robot from entering restricted areas and also improve the time that a delivery round takes by avoiding unneeded paths. To achieve this, artificial vision and transfer learning techniques are used to train the YOLO v5 model, allowing the robot to detect and recognize a type of standard visual signal that indicates restricted areas. The implementation is performed on a Raspberry Pi 4, chosen for its processing power, 8GB of RAM, and easy installation on a mobile robot. Experimental results show that the implementation of this architecture in the mobile robot achieves a detection with maximum reliability of 93% in distances of 0.6 meters between the robot camera and the signal to be detected, thus allowing safe navigation of the robot and reducing the time to complete the delivery round.