APPLYING DEEP LEARNING MODELS FOR WEAPON DETECTION IN PUBLIC ENVIRONMENTS THROUGH SURVEILLANCE CAMERAS

In the context of public security, the ability to detect weapons in real time through surveillance cameras is of utmost importance. This study focuses on applying fine-tuning techniques and data augmentation strategies to a Transformer-based model to enhance the capability of identifying weapons in public environments. The fine-tuning process optimized the model parameters, along with these augmentations, to improve weapon detection performance. Experimental results show that after fine-tuning, the model achieved an [email protected] score of up to 96.5%, representing a significant improvement in accuracy compared to previous object detection models. These results demonstrate the great potential for applying the model to real-time security surveillance systems, effectively detecting and addressing threats.