基于ELM的人体跌倒预测算法

Human fall prediction algorithm based ELM

为了应用于跌倒保护装置设计,需要设计一种跌倒保护预测算法,能够准确并快速的区分跌倒动作和正常行为动作,因此提出了基于ELM的人体跌倒预测算法。该算法通过六轴传感器芯片MPU6050提取人体各个姿态下的三相加速度和三相旋转角,通过多变量分析方法得到特征量,随后对提取的特征量进行预处理,通过滑动时间窗口对数据进行切割,对处理后的数据集进行分类标签化处理,通过标签数据集进行ELM训练测试,得到一种基于ELM的人体跌倒预测算法。通过多指标理论和传统合加速度阈值算法进行了对比评估,确定了基于ELM的人体跌倒预测算法能够在0.2s内快速预测跌倒行为,并且预测准确率能够达到97.6%,完全满足跌倒预测保护装置的应用要求,并且性能明显优于传统跌倒预测算法。

In order to be applied to the design of fall protection devices,it is necessary to design a fall protection prediction algorithm that can accurately and quickly distinguish between fall actions and normal behavior actions.Therefore,an ELM-based human fall prediction algorithm is proposed.The algorithm uses the six-axis sensor chip MPU6050 to extract the three-phase acceleration and three-phase rotation angle of the human body in each posture,and obtains the feature quantity through the multivariate analysis method,and then preprocesses the extracted feature quantity,and cuts the data through a sliding time window,The processed data set is classified and labeled,and the ELM training test is performed on the labeled data set,and an ELM-based human fall prediction algorithm is obtained.Through the comparison and evaluation of the multi-index theory and the traditional combined acceleration threshold algorithm,it is determined that the human fall prediction algorithm based on ELM can quickly predict the fall behavior within 0.2s,and the prediction accuracy can reach 97.6%,which fully meets the requirements of the fall prediction protection device.Application requirements,and performance is significantly better than traditional fall prediction algorithms.