Security, safety, and protection are significant concerns in the present day’s modern world. The increasing number of crimes in public is a very intense, severe, and consequential issue regarding the safety and security of the citizenry. Moreover, in order for a country to be capitalistically powerful, it needs to guarantee a safe and secure environment for its general public, various tourists, and investors. Closed Circuit Televisions (CCTVs) cameras are being put into service in the present day’s world in order to monitor criminal activities and for surveillance, but they are of no use without human intervention and supervision. CCTV operators would be required to monitor 20–30 screens for so many hours, resulting in a lack of concentration to monitor every screen over time. Therefore, there is a primary need for automatic control systems to ensure security. The latest work in the domain of deep learning has revealed remarkable advancements in object detection. In this paper, we have contrasted various different models of You Only Look Once (YOLO)—YOLO v7, YOLO v6, YOLO v5, and YOLO v4-DarkNet on basis of parameters like Time, Maximum a Posteriori (MAP), layers, etc. All the models have been trained on sohas-weapon-detection-yolov5 and tested, and then their results have been compared for analysis. YOLO v5, which is the quickest and most accurate model for the application, stood out as the best option among the YOLO models tested for weapon identification. Applying any of these models can help us automate surveillance systems to save lives of humans and reduce the manslaughter rate. These models could also be deployed in security robots in order to spot and detect weapons to avoid risk or assault on human life.