Detecção de disparos de armas de fogo em um ambiente urbano usando redes convolucionais em uma rede de sistemas embarcados

Abstract

This work describes the creation of a sound event detection system for firearms, to be used in a smart cities context. An embedded application was created that is capable of detecting gunshots in real time, alerts a central server and communicates through a Wi-SUN mesh network, só that it would be possible to replicate the device and use a sensor network to identify the location of the event origin. For the detection, a convolutional neural network model was used that achieved 97% accuracy when running on a desktop computer and 88% after converting the model to work with the embedded device. This project was done in collaboration with a company, that provided development resources and the structure of the network that was used.