Identificação de quedas de pessoas via aprendizagem de máquina

Abstract

The main objective in the development of this work was to train an artificial convolutional neural network, in the field of deep learning, with the purpose of identifying possible people falling on the floor, especially the elderly or those who have some kind of debilitation. This project was elaborated focusing on the verification of these through images, however, such verification only occurs after a noise sensor detects a certain level of sound power. For the effective creation of this project, the programming language Python was used, as well as the aid of the Tensorflow library, from Google company and OpenCV. Images that were freely available on the internet and free of charge were used for the creation of the database, as well as own images obtained by conventional cameras. The system works with activations in cascade, where the first step is sound detection, followed by a determination of the existence of a human performed by the library OpenCV and, if so for the existence of a person in the environment, the prediction between fall or not fall, with all the processes being made by a Raspberry integrated with a camera and a noise sensor.