A Smart Surveillance System to Detect Modern Gun Using YOLOv5 Algorithm: A Deep Learning Approach

Deep learning stands as a vibrant and highly capable technology for the precise detection of specific objects. The task of identifying objects related to criminal activities is fraught with risks and often demands substantial time investments. This study demonstrates how employing deep learning methods can effectively address these challenges. In this research, a solution leveraging the You Only Look Once (YOLO) algorithm, grounded in Convolutional Neural Networks (CNNs), is proposed. The primary objective is to identify firearms in various orientations. To train and validate the model, a substantial dataset consisting of 3000 instances is employed, encompassing a diverse set of firearm positions. Initially, YOLOv3 and Region-based Convolutional Neural Networks (R-CNN) were utilized in tandem for firearm detection. However, these models failed to yield satisfactory outcomes in terms of target object detection accuracy. Consequently, a successful introduction of the YOLOv5 model was implemented to enhance accuracy. Noteworthy advantages of this solution include reduced computational requirements, swift processing speeds, and a lightweight profile. These attributes collectively contribute to superior accuracy in object detection. The findings reveal a remarkable F1 score of 0.909, underscoring the significant potential for real-time firearm detection, as well as the recognition of criminal activities and their pertinent applications.