基于机器学习的人体虚拟惯性测量组件构建方法

摘要

In recent years,utilizing human characteristics to assist pedestrian navigation and exoskeleton robot control is one of the hot research directions in the navigation and robotic fields.Aiming at the problem that the foot mounting method of inertial measurement module cannot achieve effective measurement at high overload,a method for constructing virtual inertial measurement component of human-body based on machine learning is studied.With the data samples being taken from the simultaneous measurements of the inertial measurement components installed on the foot and the other parts of lower limbs,the construction of the virtual inertial measurement component is realized by the genetic algorithm improved error back propagation (GA-BP) neural network.In order to further improve the training effect,the training samples are screened based on gait phase detection.The joint simulation results based on Anybody and MATLAB show that the proposed method can be used to simulate the inertial measurement component's output data of the foot position by using the inertial measurement component installed near the hip joint,and also has certain adaptability to the untrained gaits.The proposed method can also be applied in the fields of pedestrian precise positioning and exoskeleton robot control.%利用人体特征辅助行人导航与外骨骼机器人控制是近年来导航与机器人领域中的热点研究方向.针对惯性测量组件足部安装方式在过载较高时无法实现有效测量的问题,研究了一种基于机器学习的人体虚拟惯性测量组件构建方法.该方法以同步采集安装于足部与下肢其他部位的惯性测量组件的输出作为数据样本,通过遗传算法改进的误差反向传播(GA-BP)神经网络实现虚拟惯性测量组件的构建.为进一步改善训练效果,采用基于步态相位检测方法对训练样本进行筛选.基于Anybody与MATLAB的联合仿真结果表明,本文所研究的方法可实现采用安装于髋关节附近位置的惯性测量组件数据,有效模拟足部位置的惯性测量组件数据.该方法对未经训练的步态也有一定的适应性.本文所研究的方法可进一步应用于行人精确定位与外骨骼机器人控制等领域.