SURVEY PAPER ON DETECTION OF WEAPON USING CNN AND YOLO

The world is becoming more unsafe day by day. The crime rate continues to rise with shootings, robberies and massacres in public areas. Due to increasing unrest everywhere, there is a great demand for safety and security. Systems such as security cameras that are equipped with real-time weapon detection technology are being implemented to help protect the safety of citizens. However, current weapons detection technologies have limitations: they are limited to detecting guns and/or knives. Criminals, on the other hand, are not limited to guns and knives. Therefore, this research is to construct an optimal neural network architecture to identify 6 commonly used weapons captured from a dataset of images. The research was conducted to develop a technology that would help arrest criminals before they commit crimes to protect the safety and security of citizens. The process starts with building an optimal Convolutional Neural Network (CNN) and an optimal transfer learning model. The validation accuracies of these models are evaluated and compared to each other. The final neural network model achieved 99.31% training accuracy and 95.55% validation accuracy, thus correctly classifying different weapon images.