A Hybrid Visual Geometry Group and Support Vector Machine Approach for Detecting Armed Individuals on Motorbikes

Abstract

The rise in crime involving armed individuals on motorbikes has become a significant concern, necessitating the development of automated security systems for effective threat detection. This study proposes a hybrid model combining VGG-16, a pre-trained convolutional neural network, and a Support Vector Machine (SVM) classifier to classify armed and unarmed persons on motorbikes. A custom dataset of 1,624 images was curated through web scraping, pre-processed, and augmented to ensure robustness. The VGG-16 model was fine-tuned to extract relevant features, while the SVM classifier categorized the images into two classes: armed and unarmed. The model achieved an overall classification accuracy of 92.62%, outperforming benchmark models. Key performance metrics confirmed its reliability and efficiency, including precision, recall, and F1-score. This hybrid approach demonstrates significant potential for real-time implementation in surveillance systems, offering an automated, efficient alternative to manual threat detection. Future work will focus on expanding the dataset and incorporating real-time detection capabilities to enhance its applicability in diverse environments.