Reforçando a segurança escolar: mitigação de ameaças por meio de visão computacional

Abstract

The worrying increase in attacks and attacks on schools has caused irreparable damage and directly affected families and the entire community involved. Given this reality, It is proposed to develop a solution to mitigate these incidents, through implementation of a Computer Vision Application Programming Interface (API). The objective is to develop and evaluate the effectiveness of this API in detecting firearms and knives in images and videos, aiming to provide an efficient tool for school security. To achieve the objectives, the You Only Look Once (YOLOv8) framework is used for weapon identification and the Django Rest Framework (DRF) for development of the API. The evaluation of the effectiveness of the model is carried out through recall tests, in which were achieved, 94.8 % for knives and 89.3 % for firearms, and accuracy was achieved 90.7 % for firearms and 95.5 % for knives, respectively, demonstrating the effectiveness of the model in identifying the classes for which it was trained. Given the results presented, it can be said that this research represents a significant advance for the school security, offering an accurate and efficient tool for early detection of threats, with the potential to save lives and protect school communities