A Smart Surveillance System for Facemask Detection Using Custom CNN—BIDNet

Abstract

Amidst the global health crisis, the importance of face masks persists as a crucial measure in mitigating contagious diseases. Wearing face masks is advised as a comprehensive strategy to lessen the transmission of infectious and airborne diseases. Face masks are required for entry into many hospitals, offices, and other establishments. The proposed Smart Surveillance System will assist in identifying individuals who are not wearing face masks to enforce this. To address this issue, we propose a face mask detection system that uses convolutional neural networks (CNN) and is reliable, lightweight, accurate, and efficient. We train and test the custom model - BIDNet - on the ‘Face Mask Detection $\sim ~12$ K Images Dataset’ (FMD-12KID) and evaluate its performance on two additional datasets: ‘Face Mask Detection Dataset’ (FMDD) and ‘Face Mask Lite Dataset’ (FMLD). An impressive 98.99% classification accuracy is achieved on FMD-12KID. BIDNet outperforms all other existing models on the comparative analysis evaluating the Accuracy, F1-Score (FS), Recall Score (RS), Precision Score (PS), Training time, Prediction Time, and Number of Parameters. For real-time capability, BIDNet is deployed on a web application where it performs extraordinarily, proving its capabilities in real-time scenarios. This research serves as a significant step toward developing intelligent systems that support public health regulations and safeguard communities in times of emergency. This is a major breakthrough in smart surveillance systems, utilizing deep learning to address critical health problems such as face mask detection. The effectiveness of CNN in practical scenarios is demonstrated by this study.

Keywords

• Binary classification
• convolutional neural networks
• deep learning
• face mask detection
• smart surveillance system