Adult Fall Prediction And Anti-Fall Device

As the population ages,protecting the safety and health of elders becomes an important issue.Fall injury is a major cause of elders’death.How to effectively prevent elders from falling is still a topic that need further investigations.To solve the above problem,this paper focuses on preventing falls and the injuries,and mainly includes the following three parts:(1)Development of a fall prediction model that is based on an inertial-measurement-unit(IMU).First,movement data of twelve subjects is recorded using an IMU during falls and daily activities in experiments.Then,the Hidden-Markov-Model(HMM)and Long-Short-Term-Memory(LSTM)learn from different features,that are extracted from the IMU data,to predict or detect falls.The performance of these obtained models are evaluated.It shows that the fall prediction model based on HMM has high accuracy for fall detection but low accuracy for fall prediction.It means that this HMM-based model is unsuitable for fall prediction.Compared to the HMM-based model,the LSTM-based model has an accuracy of 96.43%,sensitivity of 91.67%and specificity of 98.33%.Further,this model can predict a fall ahead of body collision for220.45ms in average,and thus,is suitable for fall prediction.(2)Development of a fall prediction model that is based on an IMU and insole force sensors.First a human movement measure system is developed based on an IMU and insole force sensors.This movement measure system is then used to record the movement of several subjects during fall or daily activities.The recorded data is then used to analyze the human natural motion.Combining data from the IMU and insole force sensors,several features are extracted and analyzed.An LSTM-based model is then developed based on the above features.After evaluation,is shows that the model based on both an IMU and insole force sensors has an accuracy of 98.81%,sensitivity of 95.83%and specificity of 100.00%,and can predict a fall ahead of body collision for 230.22ms in average.Compared to the LSTM-based model developed by IMU data only,this model is more suitable for fall prediction.(3)Design of a deployable fall-preventing stick.This fall-preventing stick can form a large supporting area around a person and prevent him from falling.First,the basic principle and structure of the fall-preventing stick are illustrated.Then,the kinematic,static and dynamic model of the fall-preventing stick are developed.Using these obtained models,influences of several parameters to the reaction time of the fall-preventing stick is analyzed.Simulations are performed to verified these developed models.Then experiments are carried out to test the reaction time and supporting area of the fall-preventing stick.It shows that the designed fall-preventing stick has a supporting area of 1.25m~2 in its deployed configuration and the reaction time is about 210ms.This reaction time is slightly smaller than the lead time of the LSTM fall prediction model.Finally,by combining the fall prediction model and the fall-preventing stick,experiments about predicting and preventing falls are conducted.Results show that in real-life applications,the developed fall prediction model can effectively predict a fall,and that the fall-preventing stick can form a large supporting area in a short time,and thus,preventing humans from falls.All these results validate the proposed fall prediction model and the fall-preventing stick.