多变量时序关联分析的跌倒预测算法

Fall prediction algorithm based on multivariate time series correlation analysis

为了进行跌倒预测跌倒保护装置的设计,需要一种跌倒预测算法,能够快速并准确的预测出跌倒动作。通过获取人体在站立过程中,前向弯腰、侧向弯腰、后仰、后躺和跌倒等动作的姿态数据变化情况,分析各姿态数据序列间多个变量在各种动作下的关联性,建立区分具有后向与侧向跌倒趋势和日常生活行为的判别算法。为了验证算法的有效性,利用六轴运动处理芯片MPU6050与跌倒预测处理芯片STM32F103C8T6构建了穿戴式跌倒预测触发器,通过实验表明该算法能够在0.1s内完成了跌倒预测的触发。该跌倒预测算法可以有效过滤日常生活行为和跌倒行为,并满足跌倒预测算法的准确性和快速性。

In order to design a fall protection device for fall prediction,a fall prediction algorithm is needed that can quickly and accurately predict a fall action.By obtaining the posture data changes of the human body in the process of standing,bending forward,bending laterally,leaning back,lying back and falling,analyze the correlation of multiple variables between various posture data sequences under various actions,To establish a discriminant algorithm that distinguishes between backward and lateral fall trends and daily life behaviors.In order to verify the effectiveness of the algorithm,a wearable fall prediction trigger is constructed using the six-axis motion processing chip MPU6050 and the fall prediction processing chip STM32F103C8T6.Experiments show that the algorithm can complete the fall prediction trigger within 0.1s.The fall prediction algorithm can effectively filter daily life behaviors and fall behaviors,and meet the accuracy and rapidity of the fall prediction algorithm.