Deep Learning for Smart Surveillance: Multi-Class Detection of People, Weapons, and Masks using YOLOv11

Computer vision technologies are revolutionizing the field of security, providing intelligent and automated tools for crime prevention. In this study, we propose an advanced approach for the simultaneous detection of people, weapons, and facial masks in images extracted from surveillance camera video streams. Unlike most works in the literature, which focus on static and less contextualized scenarios, our method operates on real-world, dynamic scenes using the YOLOv11 architecture. Experimental results show an accuracy of 65.35% on the test set, with a weighted precision of 71.94%, demonstrating the model's ability to reduce false positives. This work represents a concrete step towards intelligent monitoring systems, with promising improvements achievable through dataset optimization and the integration of advanced deep learning techniques, aiming to enhance public security effectiveness.