Smart Surveillance: MultiWeapon Armed Person DetectionWith Automated Threat AlertsOpen Access

Abstract

The rise in criminal activities has created an urgent demand for intelligent surveillance and command systems in law enforcement. This work introduces a deep learning–based model tailored for classifying distinct weapon categories. The model is developed on the YoloV8-CNN+LSTM framework and implemented using Keras with TensorFlow as the backend. It is trained to recognize gun, rifles, knives. A carefully curated dataset containing 5,214 images was prepared for training and evaluation. The proposed network underwent comparative testing against established models, including VGG-16, ResNet-50, and ResNet-101. Results show that the model achieved an improved accuracy of 92%, outperformingVGG-16 (89.75%), ResNet-50 (90.70%), and ResNet-101 (83.33%). These findings helps to improve proposed approach in enhancing weapon detection accuracy, thereby strengthening the capabilities of security forces in preventing crimes.