The issue of violence related to firearms has become one of the key global issues, posing a serious threat to public safety and law enforcement. Traditional surveillance systems depend heavily on human monitoring, which is time consuming, error-prone, and inefficient for large-scale applications. This study presents an automated firearm detection system using deep learning and advanced computer vision techniques to address these issues. The system leverages state-of-the-art object detection models such as Faster R-CNN and EfficientDet for the detection of firearms and human faces within surveillance footage.The detection accuracy is further enhanced by combining the outputs of multiple models using an ensemble learning approach including Non-Maximum Suppression and Weighted Box Fusion techniques. The proposed method achieves a significant improvement in the detection performance, reaching mean average precision (mAP) scores of 77.02% at IoU 0.5, 16.40% at IoU 0.75 and 29.73% across IoU thresholds from 0.5 to 0.95. Experimental results show that the ensemble model is consistently better than individual detection models in terms of accuracy and reliability.This system can analyse live CCTV footage and recorded videos to quickly identify threats related to firearms and support law enforcement agencies in taking timely preventive actions. Moreover, the approach can be generalised to analyse social media content for gun activity detection. In summary, the presented framework provides an efficient, scalable and reliable solution for intelligent surveillance and the enhancement of public safety.
REDDY et al. (Fri,) studied this question.