Desenvolvimento de um dispositivo para classificação de pisada utilizando sensores inerciais

Abstract

This work aims to develop a wearable device, based on inertial sensors and artificial neural network (ANN) for the identification of the type of step during gait, to aid diagnosis and monitoring to be performed by health professionals. The device contains a central responsible for the grouping and transmission of the signals. The two inertial modules are arranged one on the posterior surface of the calcaneus bone and the other on the gastrocnemius muscle. The reception and storage of data is performed on a computer using the Wi-Fi ™ protocol. Inertial data from the gait of 9 people with an average age of 24.8 were collected. The data were used to extract characteristics and arranged as inputs in a Multilayer Perceptron (MLP) to perform the classification of types of step. All individuals collected data under a protocol, which consists of walking 5 meters away. They were also submitted to data collection and evaluated by a professional who classified the type of step. The data were filtered, segmented, normalized and the characteristics were extracted in samples defining the database. To define the RNA configuration and the number of neurons to be used, the K-fold cross-validation method was used. The MLP with the best performance stood out the maximum, minimum and DASDV characteristic for the right foot and maximum, minimum and WL for the left foot, all of which showed an accuracy of 99.22% in the test level for both feet.