Real-time Smartphone Usage Surveillance System Based on YOLOv5

Abstract

The digital era affects students' attitudes toward utilizing applications as learning media. This phenomenon can be used to boost student achievement, but it can also have negative consequences, such as chatting while studying or cheating on school exams. To support the positive and reduce the negative impact of smartphone use, it is necessary to supervise this activity. The supervision can be done by utilizing a camera to detect a smartphone. The YOLOv5 algorithm was used, which is known for its good speed and accuracy in object detection. This smartphone detection system can be controlled, so the application is adjustable to the needs of learning activities. Collecting a dataset, annotating, training objects, writing program code, and testing the system are all stages in the development of this system. The dataset used in this research consists of 1,038 smartphone images from the internet and camera-captured images. This detection system was built to assist teachers in monitoring the use of smartphones by students. The results of this model training are 77.7% mean average precision, 93.2% precision rate, and 71.7% recall rate under varying lighting conditions.