Improved Deep Neural Network for Real-time Detection of Some Publicly Prohibited Items

Abstract

Abstract An important approach that has been put in place for the purpose of ensuring that people are secure and safe in public places is the security check mounted at the entrance of most public places. On some occasions, it is trained officers in this field that are used while in some places they are not professionally trained. For some other places, x-ray scanning machines are installed to do this task; for example, in places like airports and railway stations to help in checking for prohibited items thereby ensuring the safety of travelers and transporters. The use of these machines has helped in minimizing crime recorded in these areas significantly. However, some other places like hospitals, schools, and event centers may not have the luxury to install such devices and employ professionals that will work there, thereby exposing the people in such areas to safety threats. This raises a security concern since safety may not be guaranteed in such places. Additionally, because of the way that baggage is packed by people, some prohibited items may be smuggled into public places unnoticed even if x-ray scanners or some persons are employed to carry out security checks at the door. The tendency to perpetrate evil within the premises is possible if luggage with the prohibited item is smuggled in. It is with that in mind that we designed a real-time detection model on the basis of a deep neural network that is able to detect publicly prohibited items. We manually annotated the dataset we used and utilized the benefits of Deep Neural Networks (DNN) for the detection of the 9 classes of objects that we have in our dataset. We as well used different input sizes (416 * 416 and 608 * 608) for the training of the model and were able to compare the performance of the two different input sizes. From the result we obtained from the training, the image input size of 416 gave a better performance with an mAP of 76.75% as well as a speed of detection of 27.1 Frames per Second (FPS).