Fall Detection using Human Skeleton Features

Abstract

Falls are one of the leading causes of death and serious injury in people, especially for the elderly. In addition, falls accidents have a direct financial cost for health systems and, indirectly, for the productivity of society. Among the most important problems in fall detection systems is privacy, limitations of operating devices, and the comparison of machine learning techniques for detection. This article presents a fall detection system by means of a k-Nearest Neighbor (KNN) classifier based on camera-vision using pose detection of the human skeleton for the features extraction. The proposed method is evaluated with UP-FALL dataset, surpassing the results of other fall detection systems that use the same database. This method achieves a 98.84% accuracy and an F1-Score of 97.41%.