Design And Implementation Of Wearable Device Fall Detection System

With the continuous changing of China’s population structure and the increasing number of the elderly people,the degree of aging in China is gradually deepening.Falling problems have become one of the biggest threats to the health of the elderly,so how to ensure that the elderly can get timely assistance after falling is an important technical problem.This paper investigated the latest developments in the field of fall detection at home and abroad in detail,and chose fall detection based on wearable devices.By using Tiny Machine Learning(Tiny ML)technology,the fall detection algorithm based on machine learning and deep learning is successfully applied to embedded terminal,and a high-precision fall detection device is developed.The main work and innovation of this paper are as follows:(1)An experimental data acquisition scheme is designed for our fall detection test.A large amount of diverse dataset is used in training of machine learning algorithms.Therefore,in addition to using public data sets,we had independently designed experiments to collect our own fall data sets.Fifteen volunteers were invited to wear data collection devices to simulate the movements of daily life and falls of the elderly.The data set covered a variety of movements,which were closer to the real scene.(2)A fall detection system based on raspberry PI was designed.Using raspberry PI as a fall detection terminal device,its data does not need to be uploaded to the cloud or server,and machine learning model inference can be carried out in the local terminal,which not only reduces the energy consumed in the data transmission stage,but also avoids the data privacy leakage.According to the filtered data,the feature set for fall detection is extracted and a Support Vector Machines(SVM)algorithm is used to judge a fall event is happened or not.An improved grid search method is also used to optimize the parameters of SVM model to find the best parameter combination.(3)A fall detection system based on Tiny ML is designed.The Raspberry PI is the equivalent of a tiny computer with an operating system.While it can run machine learning algorithms,it’s too power-intensive and bulky for wearable devices.Therefore,a wearable fall detection device with low power consumption and high precision was developed by applying Tiny ML technology.Using Arduino Nano 33 BLE as a main control module of fall detection system,a PCA-ANN algorithm for fall detection was also proposed.The Tensor Flow framework is used to train an Artificial Neural Network(ANN)model,and a Principal Component Analysis(PCA)is used to carry out feature dimension reduction as the purpose for reducing the energy consumption of equipment in feature extraction stage.Finally,the trained model is transformed,compressed and deployed to a tiny Arduino terminal,which basically meets the requirements of wearable devices.