| Popis: |
According to the world health organization (WHO), the falls are the second cause of death worldwide related with unintentional injuries, affecting mainly adults over 65 years old. In this article, a portable embedded system is developed through the use of a 32-bit microcontroller and an inertial sensor, that is used for three activities of daily living (ADLs) detection: falls, walking and stationary state. In embedded systems the amount of data to be processed (time window) affects directly the response time and define the features of the hardware to be selected and used. Within a different time windows, the accelerometer signals \(a_{x} ,a_{y} ,a_{z}\) are acquired in order to calculate the acceleration module over the time. By the use of descriptive statistics, this module will provide features as: mean, median, variance, 25th percentile and asymmetry in order to be used in the Naive Bayes classifier, which was selected based on a previous work. The system has wireless communication to send information from the detected ADLs to the PC for test purposes. The device was carried by 6 people on the waist, who carried out the tests of the different ADLs. System accuracy is between 60% and 82% with processing times between 0.5–1.35 ms depending of the time window size. For the offline training of the Naive Bayes algorithm, the Data Set of the Institute of Communications and Navigations is used, corresponding to samples of 16 male and female subjects between 23 and 50 years old. |
