A Movement Decomposition And Knn-Based Fall Detection

Abstract

Aging in people makes them vulnerable to falls and this has become a universal problematic fitness issue. Literature provides several resolutions for the recognition of falls among which the wrist worn strategies are included as part of improving the output and efficacy more than that of 95%. In accordance with the theory of comfortability in the aged people, wrist is considered as the most effective area proposed for the equipment to be placed. So, the paper puts forward the idea of an apparatus to solve the mentioned issue of fall recognition. Diverse sensing units (accelerometer, gyroscope, and magnetometer), along with the indicators (acceleration, velocity, and displacement), and track apparatuses (vertical and nonvertical) are collectively used in addition to wide-ranging group of approaches involving threshold-based and machine learning theories. Therefore, it was possible to achieve superlative tactic in terms of recognizing falls. 22 individuals were chosen to learn the activities leading to fall and non-fall movements. In case of the procedures handling threshold-based idea, a maximum accuracy of 91.1% of highest exactness got accomplished in association with the acquired values of 95.8% and 86.5% in the cases of sensitivity and specificity, respectively consuming Madgwick’s decomposition. About 99.0% exactness was realized by considering the identical movement decomposition and machine learning methods during the sorting period. Sensitivity and specificity were of 100% and 97.9% respectively with the established statistics.