Real-Time Detection of Armed Individuals in Video Surveillance Using Machine Learning Techniques

Abstract

By using various methods on object detection models, a lot of research attempts to improve weapon detection. However, there is a dearth of study on using real-time surveillance cameras to identify armed individuals. The creation of algorithms to recognize individuals carrying weapons (pistols and revolvers) is the suggested remedy. The YOLOv4 model is the one we have selected to identify faces, firearms, and individuals. Then, in order to identify the armed individuals in each video frame, we extract information from YOLO pertaining to real-time movies, including bounding box coordinates, distances, and intersection regions between firearms and the individuals. Some obstacles must be overcome, such as occlusion, concealed firearms, and persons in close proximity to one another. It enables us to create and contrast various kinds of solutions. We suggested seven machine-learning models and three heuristics. The three heuristics are the principle of distances, the principle of intersections, and the principle of centers. The Random Forest Classifier, Multilayer Perceptron, k-Nearest-Neighbors, Support Vector Machine, Logistic Regression, Naive Bayes, and Gradient Boosting Classifier are other machine learning models.

Introduction

The study aims to enhance video surveillance systems to automatically detect armed individuals, focusing on pistols and revolvers, using deep learning and computer vision. The system employs the YOLOv4 object detection model trained on a custom dataset of 5,000 images, detecting three classes: handguns, individuals, and faces. Detection heuristics—center, intersection, and distance of bounding boxes—are applied to identify armed persons, improving response times and security monitoring.

The literature review highlights advancements in machine vision, 3D point cloud analysis, and anomaly detection in video surveillance. Deep learning models, particularly convolutional neural networks (CNNs), have shown state-of-the-art performance in object detection and anomaly recognition, achieving high accuracy for real-time security applications.

The proposed system includes modules for dataset loading, data augmentation, preprocessing, training, classification, and model evaluation. The CNN model is trained using an 80/20 train-test split, and results are validated using classification metrics and confusion matrices. This automated approach aims to reduce reliance on human monitoring and enhance safety in residential, workplace, and public environments.

Conclusion

For weapon (gun) detection, SSD and Faster RCNN algorithms are simulated for pre-labeled and self-generated image datasets. Both algorithms are effective and produce good results, but using them in real time requires balancing accuracy and speed. With a speed of 0.736 s/frame, the SSD algorithm provides faster performance. Faster RCNN, on the other hand, only provides 1.606 s/frame, which is not as fast as SSD. Faster RCNN provides improved accuracy of 84.6% in this regard. In contrast, SSD\'s accuracy of 73.8% is subpar when compared to RCNN\'s speed. higher RCNN offered better accuracy, but SSD\'s higher speed allowed for real-time detection. Furthermore, by using GPUs and expensive DSP and FPGA packages for training, it can be applied to larger datasets.

References

[1] \"SSD: Single Shot MultiBox Detector,\" by Wei Liu and colleagues, European Conference on Computer Vision, Volume 169, pp. 20–31 September 2017. [2] D. Erhan and colleagues, \"Deep Neural Networks for Scalable Object Detection,\" IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2014. [3] \"Anomaly Detection in Videos for Video Surveillance Applications Using Neural Networks,\" written by Ruben J. Franklin and colleagues, International Conference on Inventive Systems and Control, 2020. [4] The article \"A Review of Artificial Intelligence Methods for Data Science and Data Analytics: Applications and Research Challenges\" was written by H R Rohit and colleagues. I-SMAC (IoT in Social, Mobile, Analytics, and Cloud) had its second international conference in 2018. [5] Using a Neural Network Framework to Classify Objects in Video Records, Abhiraj Biswas et al., International Conference on Smart Systems and Inventive Technology, 2018. [6] IEEE International Conference on Intelligent Sustainable Systems, 2019; Pallavi Raj et al., \"Simulation and Performance Analysis of Feature Extraction and Matching Algorithms for Image Processing Applications.\" [7] In 2018, Mohana and colleagues presented their work, \"Simulation of Object Detection Algorithms for Video Survillance Applications,\" at the International Conference on I-SMAC (IoT in Social, Mobile, Analytics, and Cloud). [8] In 2019, Yojan Chitkara and colleagues presented their findings at the International Conference on Computing Methodologies and Communication on \"Background Modelling techniques for foreground detection and tracking using Gaussian Mixture model.\" [9] Rubner et al. (2016) presented \"A metric for distributions with applications to image databases\" at the International Conference on Computer Vision. [10] “Evaluation of Object Detection and Tracking Algorithms for Traffic Surveillance Applications with Neural Networks,\" Third International Conference, 2019 N. Jain et al.

Copyright

Copyright © 2026 P. Srinivasulu, Leela Harshini T. 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.