An Efficient LSTM and SVM based Fall Event Detection System for Elderly People

Abstract

In recent days, there is a lack of time and energy for the children to care for the elderly people, thus the elderly people monitoring system is developed to take care of them. This system is developed especially for monitoring the living conditions of elderly people in real-time.  In the future, with this proposed system one can monitor the elderly by using the alarm function. Here mainly the falling detection is done, which is one of the major issues in the elderly monitoring systems. The objective is to identify the fall event as normal and non-fall events as abnormal for monitoring elderly people. In this research paper, a deep learning mechanism is used to deduct the fall and build a fall identification system that is efficient. The Long ShortTerm Memory (LSTM) recurrent neural networks are one of the deep learning models used for feature extraction that is used for automatically identifying an elderly person’s unusual behaviors. The proposed method is tested on UR Fall, Weizmann and KTH datasets.Its performance is evaluated using Precision, Recall, F-score and the accuracy of classifications. Finally, the result is compared with other deep learning algorithms. The investigational result of proposed work proved that its supremacy by achieve 100% accuracy for the UR Fall dataset, 95% for the Weizmann dataset and 96% for the KTH dataset which is more efficient than the other existing algorithms.