Research On The Fall Prediction System Of The Elderly

Now the world is facing the serious problem of population aging. Due to decline in physical function and balance ability, poor eyesight and other reasons, the elderly are prone to fall. The fall detection system can detect falls and give an alarm, which will reduce the time of waiting for rescue when the elderly fall, reduce the harm and medical expenses caused by falls and enhance the elderly’s confidence of living independently.After analysing the current fall detection methods, a fall detection method based on wearable device is adopted. The innovation of this paper is that acceleration and angular features are combined and support vector machine(SVM), a kind of machine learning algorithms, is used as the classification algorithm to detect falls prior to the fall impact. The fall prediction problem is treated as a binary classification problem and the classification model is built to realize fall prediction. Then remote fall alarm and fall protection device is studied.Firstly, various fall prediction algorithms are described and the algorithm of SVM is determined as the classification algorithm. Then the theory of SVM is expounded and the process of realizing the algorithm of SVM and the method of parameter optimization is described. Then the data acquisition equipment is worn at the waist of human body and the human body 3D coordinate system is established to acquire the data of 3 axis acceleration, 3 axis angular velocity and 3 axis magnetic field. Four kinds of fall behaviors, ten kinds of activities of daily living(ADL) and statistics of acceleration and angular velocity are defined. The characteristics of acceleration, angular velocity and attitude angle during different behavior process are collected and analyzed and it is found that using these characteristics to detect falls is strongly feasible.Based on the analysis of characteristics, the characteristics of human behavior are extracted. According to the change law of the statistics of acceleration and angular velocity, 4 acceleration features and 4 angular velocity features are extracted in the time domain. The attitude angle is calculated in the inertial navigation system and the feature of attitude angle is extracted. In all, 9 features are extracted. Then the samples of fall and ADL are collected to build the training set and the test set and the variables in the process of collecting samples are discussed. The sequential forward selection method and sequential backward selection method are used to obtain the optimal combination of features, which contains 3 features and ultimately the optimal fall prediction model is build. The fall detection rate and the ADL detection rate is 100% and 100%, respectively. The average lead time is 291 ms. All these can validate the effectiveness and feasibility of the proposed fall detection algorithm in this paper.Finally, the remote alarm of fall and fall protection device are studied. The SIM900 A module is used to realize fall message alarm. A manual airbag device is transformed into an automatic airbag device. Two different connecting rod mechanisms are designed to connect the motor and the air bag device and through the experiment the connecting rod mechanism with less time consuming is selected.The time required to start the airbag device is about 133 ms through the experiment so under the current optimal fall detection classification model, the time left to inflate the airbag is about 158 ms. According to the calculation, it is known that the time of complete release of the cylinder is about 452 ms in the condition of the existing cylinder and it is verified by experiments. Then the cylinder specification is calculated to satisfy the 158 ms inflatable time. Under the condition of the existing cylinder, the fall protection device is tested by experiments and the results indicate that the fall protection device can be started before people falling to the ground.