The detection of weapons supported by deep learning algorithms is a research topic and application in the domains of public safety, law enforcement, security, and defense. The motivations lie in the increase in crimes with the use of weapons, near schools or other institutions. Therefore, closed circuit television (CCTV) systems play a key role in security monitoring, however, some gaps are enumerated, such as the requirement of human labor to supervise the area, which is a repetitive and time-consuming task. Previous works have faced several hurdles, especially in instances of poor weapon visibility and low-quality imagery. This approach aims to address the enumerated challenges by improving the capability for real-time detection of weapons within CCTV systems. The objective is to create an end-to-end system, named SafeGuard, that assists law enforcement in pinpointing criminal behavior on online and offline video footage. This approach aims to improve model detection accuracy, making them more effective and resulting in better security. The preliminary metrics results of the proposed machine learning model achieved an Accuracy of 0.91.