Real-Time Firearm Detection System Utilizing Deep Learning and Super-Resolution CNNs

Human monitoring and involvement are still required at the current level of surveillance and control systems. Deep learning improvements in recent years have shown tremendous progress in object recognition and categorization. This study provides a unique approach for automatically recognizing weapons in videos that may be used for both surveillance and control. Initially, researchers produce a variety of datasets that include a variety of image types, such as blurred photos, night scenes, dark backgrounds, and visuals with complicated backgrounds. The combination of YOLOv4 and SRCNN-based models (referred to as SRYolo) trained on newly collected datasets yields the most promising results. The model achieved a precision score of 85.22% and an F1-score of 80.98%, making it effective as a weapon detector.