Copyright © 2025 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{163419,
        author = {S.TEJASWI(ASSISTANT PROFESSOR) and D.Vikas and k.Tarani and N.Aswan and S.Apparao},
        title = {Deep Learning-Based Real-time Weapon Detection in CCTV Surveillance Footage},
        journal = {International Journal of Innovative Research in Technology},
        year = {},
        volume = {10},
        number = {11},
        pages = {1153-1159},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=163419},
        abstract = {In today's modern world, ensuring security and safety is paramount for a country's economic strength and attracting investors and tourists. While Closed Circuit Television (CCTV) cameras are commonly used for surveillance, they still rely on human supervision to detect illegal activities such as robberies. The challenge remains in developing a system that can automatically detect such activities, especially weapon-related threats, in real time despite advancements in deep learning algorithms, hardware processing speed, and camera technology.
This work focuses on enhancing security using CCTV footage to detect harmful weapons by leveraging state-of- the-art open-source deep learning algorithms. The approach involves binary classification, with pistols as the reference class, and introduces the concept of including relevant confusion objects to reduce false positives and false negatives. Due to the lack of a standard dataset for real-time scenarios, a custom dataset was created using weapon photos from various sources, including personal cameras, internet images, YouTube CCTV videos, GitHub repositories, data from the University of Granada, and the Internet Movies Firearms Database (IMFDB).
Two main approaches were employed: sliding window/classification and region proposal/object detection. Several deep learning algorithms, including VGG16, Inception-V3, Inception-ResnetV2, SSDMobileNetV1, Faster-RCNN Inception-ResnetV2 (FRIRv2), YOLOv3, and YOLOv4, were tested based on precision and recall, which are more crucial metrics than accuracy for object detection tasks.
Among these algorithms, YOLOv4 demonstrated superior performance, achieving an F1-score of 91% and a mean average precision of 91.73%, surpassing previous benchmarks. This highlights its effectiveness in accurately detecting harmful weapons in CCTV footage, contributing significantly to enhancing security measures.
},
        keywords = {},
        month = {},
        }