Human activity recognition using inertial sensors and machine learning

• Grant number: 18/06463-6
• Support type: Scholarships in Brazil - Scientific Initiation
• Effective date (Start): June 01, 2018
• Effective date (End): December 31, 2018
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Theory of Computation
• Principal researcher: Roseli Aparecida Francelin Romero
• Grantee: Felipe Aparecido Garcia
• Home Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil

Abstract

In this project, we aim to recognize human activity applying different classification techniques to public databases with information on various tasks carried out by volunteers. The main techniques to be used will be based on Deep Learning, in particular architectures of recurrent neural networks, such as Long Short-Term Memory (LSTM). The main datasets to be worked will be SBHAR, which has inertial data of smartphones fixed to the waist of the users, and PAMAP2, which consists of data from wearable sensors distributed by the body of users and cardiac monitors used during the proposed tasks. In addition, it is intended to apply to the same databases other recognition techniques for benchmarking. Among them, the main one will be the Hidden Markov Models. It is intended, therefore, to determine the best technique, among those used, to classify the data worked, initially limited to inertial data. These recognition techniques will be applied to smart-homes. (AU)