Temporal Approaches for Human Activity Recognition Using Inertial Sensors
Felipe Aparecido Garcia, Caetano Mazzoni Ranieri, Roseli Aparecida Francelin Romero · 2019
Human Activity Recognition (HAR) involves classifying one person's activity based on sensor data. In this work, inertial data, collected mainly from wearable sensors, are used to recognize HAR by using Convolutional Neural Networks (CNN) to extract features from the raw sensor data. Additionally, Temporal Convolutional Networks (TCN) are applied to classify the extracted features, comparing the overall performance of those layers with Long Short-Term Memory (LSTM) recurrent neural network layers. Several experiments are performed and the results show that TCN based architectures are able to outperform LSTM based architectures in sequence modeling.