AI Powered Weapon Detection for Public Safety

Abstract

With public spaces facing a growing number of security challenges, there\'s an urgent push for smarter surveillance systems .By integrating Artificial Intelligence and computer vision, we now have a powerful means to detect weapons and prevent threats in real time. This survey paper presents a A detailed overview of the available research and practices deep learning-based techniques, with a particular emphasis on the YOLO family of models, including the latest YOLOv8. These models demonstrate strong capabilities in identifying weapons such as guns and knives from live video streams with high precision and speed. The survey highlights how modern frameworks can blend effortlessly into the existing system into current surveillance infrastructures to deliver scalable, low- latency, with instant observation and feedback mechanisms solutions. Furthermore, the paper discusses datasets, evaluation metrics, application domains, and key challenges such as privacy, false alarms, and scalability. By synthesizing findings across drawings from the latest research, this survey explores key perspectives on. current state of AI- driven weapon detection and propose next steps to advance understanding in the field for building safer public environments, including schools, airports, and government facilities.

Introduction

Recent advances in deep learning have enabled real-time detection of weapons, such as guns and knives, from live video streams with high accuracy. Object detection frameworks like Faster R-CNN, SSD, and especially the YOLO family (YOLOv3–YOLOv8) balance speed and precision, making them suitable for security-sensitive environments. Surveys of the field highlight key priorities including scalability, low-light performance, false alarm reduction, and integration with existing CCTV infrastructure.

Studies using YOLOv5, YOLOv7, and YOLOv8, often supported by CNNs and transfer learning, demonstrate real-time detection with precision and recall above 94–95%, analyzing live feeds within 1–1.5 seconds. Methodologies typically involve large, diverse datasets, preprocessing, and post-processing techniques like non-max suppression to optimize accuracy and reliability.

While AI-powered detection significantly improves response times compared to manual monitoring or traditional security methods, challenges remain in handling occlusion, cluttered environments, low lighting, and false positives. Ethical deployment, privacy considerations, and multimodal enhancements (e.g., thermal imaging) are emphasized for practical use. Overall, these systems show strong potential for enhancing public safety by automating threat identification in high-risk environments.

Conclusion

This research set out to explore the potential of AI- powered weapon detection as a means of strengthening public safety in high-risk environments. Utilizing architectures like CNNs for spatial feature extraction and LSTMs for sequential data modeling, the system achieves enhanced predictive accuracy YOLOv5, YOLOv7, and YOLOv8, By integrating multiple deep learning architectures, the system achieves enhanced performance across complex tasks with Convolutional Neural Networks, the system was able to identify weapons such as guns and knives in real time while maintaining strong accuracy and efficiency. In contrast to conventional surveillance approaches dependent on manual monitoring and predefined rules heavily on human monitoring, the proposed framework processes video feeds automatically, significantly reducing response times and enhancing situational awareness. Its scalability and adaptability to enhance its adaptability and integration within operational frameworks existing CCTV infrastructure, making it a practical and impactful solution for modern security challenges. At the time, of the study acknowledges certain limitations that must be rigorously examined to ensure system robustness before large-scale deployment. Detection performance can still be hindered by occlusions, poor lighting, and visually similar non- weapon objects, occasionally leading to false alarms. These challenges open Further work may focus on curating enriched datasets that capture broader environmental and contextual variability, applying domain adaptation techniques, and incorporating multimodal sensing with thermal or infrared cameras. Ethical considerations, including privacy and responsible use of surveillance technologies, must also remain at the forefront. In conclusion, AI-powered weapon detection presents a promising direction for enhancing security systems, and with continued improvements, it has the potential to become a cornerstone of intelligent surveillance in public spaces.

References

[1] Bochkovskiy, A., Wang, C. Y., & Liao, H. Y. M. (2020). YOLOv4: Optimal speed and accuracy of object detection. arXiv preprint arXiv:2004.10934. [2] Reza, S, & Mahmud, S. (2021). Real-time object detection system for embedded devices using YOLOv5 IEEEAcess,9,150900150910. doi:10.1109/ACESS.20221.3125234. [3] A. Thakur, A. Shrivastav, R. Sharma, T. Kumar, and K. Puri, “AI-powered weapon detection for enhanced public security,” arXiv preprint arXiv:2410.19862, 2024. [4] S. A. Begum, K. S. Reddy, N. Subbulakshmi, S. B. Jabiulla, P. Sreenivasulu, and G. Raju, “AI-powered weapon detection for enhanced public security,” in Proc. 5th Int. Conf. Data Intelligence and Cognitive Informatics (ICDICI), 2024, pp. 1385–1390, doi: 10.1109/ICDICI62993.2024.10810984. [5] U. Gupta, “Weapon detection and alerting system using deep learning,” Int. Res. J. Modernization Eng., Technol. Sci. (IRJMETS), vol. 7, no. 4, pp. 2757–2761, Apr. 2025. [6] P. V. Gore and M. D. Katkar, “Real-time weapon detection using deep learning and computer vision,” Int. J. Res. Anal. Rev. (IJRAR), vol. 11, no. 4, pp. 365–368, Nov. 2024. [7] V. S. Padmini, S. Ashfaquddin, U. Kodge, and S. M. A. Quadri, “Weapon detection using artificial intelligence & deep learning for security applications,” J. Emerg. Technol. Innov. Res. (JETIR), vol. 12, no. 1, pp. 197–201, Jan. 2025. [8] Uddin, M., & Hassan, M. (2019). Weapon detection in CCTV using deep learning. IEEE International Conference Image Processing (ICIP),42984302doi:10.1109/ICIP.201 9.8880898

Copyright

Copyright © 2025 Dr. Anitha T G, S Pavan, Yashavanth , Yashas H J, Tejaswi P Achar. This 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.