Personalized Fall Detection Based On Feature Analysis

As the aging process quickened, the world’s population presents three features of the old aging, aging population, and empty-nested. This makes the state and society to face great pressure, and cause the attention of the whole international society. Fall is an often happened accident in elderly people, therefore the elderly fall detection has become a hot spot at home and abroad in recent years, having very high research value and application significance. Therefore, the development for the elderly fall detection system which can accurately distinguish the elderly fall events from normal life behavior without affecting the elderly people’s normal life, can effectively improve the life quality of the elderly people. However, there are still some problems in the study of the elderly fall detection:one problem is that unified model and same threshold are used for all of people in the case of not take into account the individual differences between people; the other problem is that it can lead high misjudgment of falls and similar motions.This paper aims to study the personalization features of fall, verifying the effectiveness of the proposed feature through experiments. On this basis, in order to further improve the detection accuracy, we choose different features for different people and setting up personalized classification model. Paper’s main works are summarized as follows:(1) We analyze the diversity of movement posture and range and the part which ha server acceleration change in the process of fall for different people, so for different individuals we extract features from different parts for fall detection in order to solve the problem of feature variability between individuals.(2) More useful information is needed due to the complication of human movement. This paper uses the method that more sensors are wear in multiple parts of human body, and then we combine and compute the information. After that, using feature analysis and selection algorithm, it is verified the existence of individual differences. Therefore, according to the specific people different characteristics should be extracted.(3) For specific individuals, universal model have the disadvantages of low detection rate and high false alarm rate. In this paper, by using Bayesian framework for feature selection and machine learning algorithm, it is verified that using the personalized classification model which is built basing on the training data of specific people can get a higher detection rate, and a lower false alarm rate.