Abstract:

Violence remains a pervasive issue in modern society, with firearm-related incidents posing a significant threat to public safety and challenging law enforcement agencies worldwide. Despite advancements in civilization, firearms continue to be a common means of violence, particularly in urban and semi-urban areas. While CCTV-based surveillance systems are widely deployed for crime prevention and monitoring, traditional human-based monitoring is both resource-intensive and prone to errors. To address these limitations, this project explores the application of deep learning techniques for automated firearm detection, specifically focusing on guns. The system leverages advanced object detection methods such as Faster Region-Based Convolutional Neural Networks (Faster R-CNN) and EfficientDet architectures to accurately identify guns and human faces in surveillance footage. Additionally, ensemble methods—including Non-Maximum Suppression, Non-Maximum Weighted, and Weighted Box Fusion—are implemented at the post-processing stage to enhance detection accuracy. Through comprehensive comparative analysis, this study demonstrates how integrating multiple detection approaches improves overall performance. The proposed system aims to assist law enforcement by enabling quicker response and proactive intervention, thereby contributing to public safety and the reduction of firearm-related violence.

Description:

.