三軸加速度計による高齢者を対象とした転倒検知アルゴリズムに関する研究

抄録

 Since Japan has a very aged society and the proportion of elderly people is increasing, falling down accidents of elderly people are increasing year by year. Since falling down accidents cause bedridden, it is one of the causes threatening the daily lives of elderly people. Therefore, early detection at the time of falling down is a big problem in Japan.

 In this research, we developed an algorithm which can detect falling down of the elderly accurately, with the aim of watching the elderly. For detection of falling down, a triaxial accelerometer which is often used for wearable devices was used. In creating the algorithm, we analyzed physical information (such as age, weight and height) and waveform of acceleration when falling down. In addition, we proposed to define the threshold of the maximum combined acceleration when falling down. The research procedure is shown below.

 First of all, we investigated the acceleration of young people and the elderly in their daily lives. As a result, it was found that the maximum combined acceleration is different between young people and elderly people.
 Next, the falling test was conducted by 5 subjects aged between 20 and 70 years old, and acceleration data at the time of falling down, 72 times was obtained. Next, based on this experiment, an algorithm for detecting falling down was created. The algorithm was constructed based on the following four acceleration characteristics:
 (1) The maximum combined acceleration when falling down is 3.5 G or more.
 (2) The posture just before falling down is standing.
 (3) The angle of the body before falling down and after falling down is 70 degrees or more.
 (4) After falling down, the condition of falling down continues for more than 2 seconds.
 When all of the above four points are satisfied, it is judged to be “falling down”.
 Next, the accuracy of the created algorithm was verified. The accuracy verification checked whether or not it was judged to be a fall when actually falling down and whether or not it was judged to be a fall though it was not falling down.
 As a result, using the created algorithm, it was possible to detect 70 falls over 72 falling down data. Moreover, in the verification of false detection related to a fall in everyday life, only 2 out of 17 days were mistakenly detected as falling down.
 From the above, it was confirmed that falls can be detected with high precision by the created algorithm.
 The fall detection algorithm created in this research is useful for early detection for the emergency situation of elderly people. Therefore, it is possible to secure the safety of the lives of elderly people and to prevent lonely death caused by falling down of elderly people living alone.