Publication

Abstract : The number of criminal activities has been increasing rapidly in recent years. Violence committed by brandishing a knife, gun, or other weapons results in an unsafe and unsecure environment. To guarantee a safe and secure environment, smart video surveillance systems are becoming very prominent. As demand for safety and security increases, intelligent monitoring relies more on video surveillance systems that can identify and comprehend the crime scene and abnormal instances. This study proposes a computer vision-based method for weapon detection capable of recognizing knives and pistols for real-time security surveillance. Footages obtained from Closed Circuit Televisions (CCTVs) serve as the data for the neural network model. The EfficientDet model from Tensorflow2 Model Zoo is used in the proposed methodology to identify and detect weapons. The EfficientDet model has improved precision and accuracy while reducing the real-world constraints of model size and latency.