Classificação de Atividades Humanas com Redes Neurais Artificiais com Processamento Temporal


This work aims at investigating the adoption of neural networks with temporal processing to detect the type of activity developed by different subjects in an uncontrolled environment using data from accelerometer and gyroscope located in their upperarm and forearm. The results reported in the literature describe an accuracy of 75% for neural networks in this scenario. Aiming at designing neural networks more suited for this task, time delay neural networks were considered and their resulting average accuracy was equal to 86%. The models obtained can collaborate in the development of solutions to wearable devices, for example, encouraging users to become more active.

In: IV Escola Regional de Informática Norte I. Manaus, Amazonas.