Detector de Personas con Armas de Fuego a Partir de un Sistema de Visión Artificial Basado en el Análisis de Posturas Corporales

Abstract

In the social environment, every day it is appreciated how insecurity is an overwhelming evil, both for people and for the community in general. The most used objects to perpetrate this type of criminal acts are firearms. With this project, it is proposed to implement a system based on artificial vision that can, through Deep Learning algorithms and body positioning tools such as OpenPose, recognize a person with a firearm. The classification of objects or firearms is implemented with convolutional neural networks (CNN). To make this image processing faster and more effective, several development techniques will be used on Google Colab, Jupyter Lab, Oracle Virtual Machine and Git Bash, also taking advantage of the use of an NVIDIA RTX A5000 model GPU card for faster execution of each of the steps of the proposed development. The proposed project phases are five: database development, model training, validation, OpenPose implementation, model results. It also has several sub-phases that allow the efficient implementation of the project. Finally, the system made it possible to confirm the dangerous activity with a firearm through detections obtained through videos in real time.