Detection and classification of firearms applied to entertainment media

Resumo

Entertainment media have evolved considerably over the years. With this, the immersion and reality in
the production of films, series and electronic games has increased. By improving this virtual reality, there are side
effects in the production of content aimed exclusively at the younger audience, which, depending on the content,
usually has an excessive load of violence, especially content related to firearms. Although the state of the art
already has a significant advance in object detection through deep learning algorithms, real-time weapons detection
is still a challenge. The detection of two classes of firearms was introduced to this work: handgun and heavygun.
A dataset containing 2,000 images was used and another 2,000 were collected from various entertainment media
and later annotated. The YoloV5 algorithm was used in this research, since it is already a consolidated model
among researchers in the area. The analysis of the result is based on the exposure time in which firearms were
exposed in entertainment content.