Módulo de detecção de quedas em cadeiras de rodas

Abstract

Mobility has always been a challenge for people who depend on wheelchairs to get around. A simple fall of the wheelchair can cause irreparable physical damage if there is no quick rescue. Aiming to propose a solution capable of facilitating mobility safety for wheelchair users, this work presents a prototype capable of detecting falls in a wheelchair and alert the event via text message to a cell phone. The detection of falls is done by an embedded system connected to accelerometer and gyro sensors, which will have their data real-time collected and processed by algorithms. These algorithms are a threshold of values, which must be established, and also by a machine learning algorithm, which implements k-Nearest Neighbors to perform the task of detecting falls. Prior to this real-time implementation, a fall and non-fall database was formed. This database was used to build kNN training set and then k-folds method was applied to this database. System’s final tests were performed on flat floor and irregular floor. System’s accuracy values are 96.9% for flat floor and 93.0% for irregular floor.