Skeleton based efficient fall detection

Abstract

Fall is one of the most important causes of death and injury for the elderly. Real time fall detection has great importance for the safety of the elderly. In this work, we presented a new method for fall detection that is based on Kinect skeleton data. Three-dimensional (3d) skeleton data of FUKinect-Fall dataset, which was constructed from 21 volunteers, is firstly reduced to two different (xy and zy) two-dimensional (2d) datasets. Then, coded regions on nested circles are constructed that are located on the determined reference joint. Thus for each action, the average of the coded 19 skeleton points during the action, are used as features. The obtained feature set is classified by using k-Nearest Neighborhood (k-NN) and Support Vector machine (SVM) classifiers. Experimental results show that fall detection is achieved with 97.08% accuracy.