计算机科学
卷积神经网络
执法
实时计算
目标检测
人工智能
模拟
模式识别(心理学)
政治学
法学
作者
Divyanka Thakur,Priya Pal,Amogh Jadhav,Numaira Kable,V Bhagyalakshmi,Sonali Deshpande
标识
DOI:10.1109/nmitcon58196.2023.10275958
摘要
Helmet and vest detection systems ensure worker safety in high-risk professions like construction, mining, and law enforcement. The latest version of the You Only Look Once (YOLO) object detection model, YOLOv8, offers significant advancements over its predecessors. YOLOv8 utilizes a fully convolutional neural network, comprising two major components: the backbone and the head, incorporating a self-attention mechanism to identify objects at various scales in the image. This model eliminates anchors, reducing box predictions and speeding up post-processing phases. Among the YOLOv8 family, YOLOv8-s stands out as the smallest and fastest model, with a size of 21 MB and a single epoch runtime of only 11 seconds. In this study, the YOLOv8-s model achieved an impressive 99.5% mean average precision (mAP) accuracy rate in real time. By implementing this system, the construction industry can significantly enhance safety practices through efficient and reliable monitoring of personal protective equipment usage at construction sites.
科研通智能强力驱动
Strongly Powered by AbleSci AI